Essentials of Metaheuristics by Sean Luke

Essentials of Metaheuristics
A Set of Undergraduate Lecture Notes by
Sean Luke
Department of Computer Science
George Mason University
First Edition
(Rev C)
Online Version 1.1
January, 2011
Figure 0 The Mona Lisa, esti-

mated with the (5 + 1) Evolution
Strategy. The objective is to find a
set of fifty polygons which most
closely approximates the original
image. After Roger Alsing.
Copyright 2009 by Sean Luke. ISBN: 978-0-557-14859-2
Thanks to Carlotta Domeniconi, Kenneth De Jong, John Grefenstette, Christopher Vo, Joseph Harrison,
Keith Sullivan, Brian Hrolenok, Bill Langdon, R. Paul Wiegand, Brian Absetz, Jason Branly, Jack Compton,
Stephen Donnelly, William Haddon, Beenish Jamil, Eric Kangas, James O’Beirne, Peshal Rupakheti, Nicholas
Payette, Lee Spector, “Markus”, Don Miner, Brian Ross, Mike Fadock, Ken Oksanen, Asger Ottar Alstrup,
Joerg Heitkoetter, Don Sofge, Akhil Shashidhar, Jeff Bassett, Guillermo Calderón-Meza, Hans-Paul Schwefel,
Pablo Moscato, Mark Coletti, Yuri Tsoy, Faisal Abidi, Ivan Krasilnikov, Yow Tzu Lim, Uday Kamath, Murilo
Pontes, Rasmus Fonseca, Ian Barfield, Forrest Stonedahl, Muhammad Iqbal, Gabriel Balan, Joseph Zelibor,
Daniel Carrera, Maximilian Ernestus, Arcadio Rubio Garcia, Kevin Molloy, Petr Pos̆ı́k, Keith Sullivan, Brian
Olson, Matthew Molineaux, and Vittorio Ziparo.
Get the latest version of this document or suggest improvements here:

http://cs.gmu.edu/∼sean/book/metaheuristics/
Cite this document as:

Sean Luke, 2009, Essentials of Metaheuristics, Lulu,
available at http://cs.gmu.edu/∼sean/book/metaheuristics/
Always include the URL, as this book is primarily found online. Do not include edition or version numbers
unless you must, as Citeseer and Google Scholar may treat each (oft-changing) edition as a different book.
BibTEX: @Book{ Luke2009Metaheuristics,

author = { Sean Luke },
title = { Essentials of Metaheuristics},
year = { 2009 },
publisher = { Lulu },
note = { Available for free at http://cs.gmu.edu/$\sim$sean/book/metaheuristics/ }
}
This document is licensed under the Creative Commons Attribution-No Derivative Works
3.0 United States License, except for those portions of the work licensed differently as described in the next
section. To view a copy of this license, visit http://creativecommons.org/licenses/by-nd/3.0/us/ or send a
letter to Creative Commons, 171 Second Street, Suite 300, San Francisco, California, 94105, USA. A quick
license summary:
• You are free to redistribute this document.
• You may not modify, transform, translate, or build upon the document except for personal use.
• You must maintain the author’s attribution with the document at all times.
• You may not use the attribution to imply that the author endorses you or your document use.
This summary is just informational: if there is any conflict in interpretation between the summary and the
actual license, the actual license always takes precedence.
Certain art and text is not mine. Figure 42 is copyright 2008 by Oskar Sigvardsson,
and is distributed under the Creative Commons Attribution 3.0 License. Figure 32 is by Wikipedia
User “Solkoll” and is in the public domain. The top Mona Lisa (in Figure 0) is from Wikipedia and is
in the public domain. The bottom Mona Lisa is mine but is inspired by Roger Alsing’s method (see
http://rogeralsing.com/2008/12/07/genetic-programming-evolution-of-mona-lisa/). Note to Roger: it’s
not Genetic Programming. The data in Table 4 is from the NIST/SEMATECH e-Handbook of Statistical Methods,
http://www.itl.nist.gov/div898/handbook/ and is in the public domain.
0
Contents
List of Algorithms 4
0 Introduction 7
0.1 What is a Metaheuristic? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.2 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
0.3 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1 Gradient-based Optimization 11
2 Single-State Methods Depends on Section 1 15

2.1 Hill-Climbing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 The Meaning of Tweak . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Single-State Global Optimization Algorithms . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Adjusting the Modification Procedure: (1+1), (1+λ), and (1, λ) . . . . . . . . . . . . . 21
2.4 Simulated Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5 Tabu Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6 Iterated Local Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3 Population Methods Depends on Section 2 29

3.1 Evolution Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.1 Mutation and Evolutionary Programming . . . . . . . . . . . . . . . . . . . . 33
3.2 The Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Crossover and Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.2 More Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.3 Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 Exploitative Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3.1 Elitism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3.2 The Steady-State Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3.3 The Tree-Style Genetic Programming Pipeline . . . . . . . . . . . . . . . . . . 46
3.3.4 Hybrid Evolutionary and Hill-Climbing Algorithms . . . . . . . . . . . . . . 47
3.3.5 Scatter Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4 Differential Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5 Variations on the Theme of Exploitation versus Exploration . . . . . . . . . . . . . . . 52
3.6 Particle Swarm Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4 Representation Depends on Sections 2 and 3 57

4.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.1.1 Initialization and Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.1.2 Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.1.3 Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Direct Encoded Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2.1 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2.2 Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2.3 Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
1
4.3 Trees and Genetic Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3.1 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3.2 Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3.3 Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3.4 Forests and Automatically Defined Functions . . . . . . . . . . . . . . . . . . 76
4.3.5 Strongly-Typed Genetic Programming . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.6 Cellular Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.7 Stack Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4 Lists, Machine-Language Genetic Programming, and Grammatical Evolution . . . . 80
4.4.1 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.2 Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.3 Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.5 Rulesets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.5.1 State-Action Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.5.2 Production Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5.3 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.5.4 Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.5.5 Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.6 Bloat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5 Parallel Methods Depends on Section 3 93

5.1 Multiple Threads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.2 Island Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.3 Master-Slave Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.4 Spatially Embedded Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6 Coevolution Depends on Section 3 103

6.1 1-Population Competitive Coevolution . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.1.1 Relative Internal Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . 107
6.2 2-Population Competitive Coevolution . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.3 N-Population Cooperative Coevolution . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.4 Niching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.4.1 Fitness Sharing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.4.2 Crowding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7 Multiobjective Optimization Depends on Section 3 127

7.1 Naive Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.2 Non-Dominated Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.3 Pareto Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
8 Combinatorial Optimization Depends on Sections 2 and 3 141

8.1 General-Purpose Optimization and Hard Constraints . . . . . . . . . . . . . . . . . . 142
8.2 Greedy Randomized Adaptive Search Procedures . . . . . . . . . . . . . . . . . . . . 145
8.3 Ant Colony Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
8.3.1 The Ant System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
8.3.2 The Ant Colony System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
8.4 Guided Local Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
2
9 Optimization by Model Fitting Depends on Sections 3 and 4 155
9.1 Model Fitting by Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
9.2 Model Fitting with a Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
9.2.1 Univariate Estimation of Distribution Algorithms . . . . . . . . . . . . . . . . 161
9.2.2 Multivariate Estimation of Distribution Algorithms . . . . . . . . . . . . . . . 165
10 Policy Optimization Depends on Sections 3 and 4 167

10.1 Reinforcement Learning: Dense Policy Optimization . . . . . . . . . . . . . . . . . . . 167
10.1.1 Q-Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.2 Sparse Stochastic Policy Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
10.2.1 Rule Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
10.3 Pitt Approach Rule Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
10.4 Michigan Approach Learning Classifier Systems . . . . . . . . . . . . . . . . . . . . . 185
10.5 Is this Genetic Programming? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
11 Miscellany Depends on Section 4 195

11.1 Experimental Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
11.1.1 Random Number Generators, Replicability, and Duplicability . . . . . . . . . 195
11.1.2 Comparing Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
11.2 Simple Test Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
11.2.1 Boolean Vector Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
11.2.2 Real-Valued Vector Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
11.2.3 Multiobjective Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
11.2.4 Genetic Programming Problems . . . . . . . . . . . . . . . . . . . . . . . . . . 209
11.3 Where to Go Next . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
11.3.1 Bibliographies, Surveys, and Websites . . . . . . . . . . . . . . . . . . . . . . . 213
11.3.2 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
11.3.3 Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
11.3.4 Conferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
11.3.5 Journals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
11.3.6 Email Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
11.4 Example Course Syllabi for the Text . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Errata 221
Index 229
3
List of Algorithms
0 Bubble Sort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1 Gradient Ascent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Newton’s Method (Adapted for Maximum Finding) . . . . . . . . . . . . . . . . . . . . . 12
3 Newton’s Method with Restarts (One Dimensional Version) . . . . . . . . . . . . . . . . 13
4 Hill-Climbing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5 Steepest Ascent Hill-Climbing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
6 Steepest Ascent Hill-Climbing With Replacement . . . . . . . . . . . . . . . . . . . . . . 16
7 Generate a Random Real-Valued Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
8 Bounded Uniform Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
9 Random Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
10 Hill-Climbing with Random Restarts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
11 Gaussian Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
12 Sample from the Gaussian Distribution (Box-Muller-Marsaglia Polar Method) . . . . . . 22
13 Simulated Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
14 Tabu Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
15 Feature-based Tabu Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
16 Iterated Local Search (ILS) with Random Restarts . . . . . . . . . . . . . . . . . . . . . . 27
17 An Abstract Generational Evolutionary Algorithm (EA) . . . . . . . . . . . . . . . . . . . 30
18 The (µ, λ) Evolution Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
19 The (µ + λ) Evolution Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
20 The Genetic Algorithm (GA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
21 Generate a Random Bit-Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
22 Bit-Flip Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
23 One-Point Crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
24 Two-Point Crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
25 Uniform Crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
26 Randomly Shuffle a Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
27 Uniform Crossover among K Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
28 Line Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
29 Intermediate Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
30 Fitness-Proportionate Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
31 Stochastic Universal Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
32 Tournament Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
33 The Genetic Algorithm with Elitism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
34 The Steady-State Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
35 The Genetic Algorithm (Tree-Style Genetic Programming Pipeline) . . . . . . . . . . . . 47
36 An Abstract Hybrid Evolutionary and Hill-Climbing Algorithm . . . . . . . . . . . . . . 48
37 A Simplified Scatter Search with Path Relinking . . . . . . . . . . . . . . . . . . . . . . . 50
38 Differential Evolution (DE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
39 Particle Swarm Optimization (PSO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
40 A Gray Coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
41 Integer Randomization Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
42 Random Walk Mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
43 Line Recombination for Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4
44 Intermediate Recombination for Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
45 Gaussian Convolution Respecting Zeros . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
46 Sample from the Geometric Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
47 Build A Simple Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
48 Build a Simple Directed Acyclic Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
49 Select a Subset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
50 Select a Subset (Second Technique) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
51 Select a Subgraph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
52 Randomly Merge One Graph Into Another . . . . . . . . . . . . . . . . . . . . . . . . . . 69
53 The Grow Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
54 The Full Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
55 The Ramped Half-and-Half Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
56 The PTC2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
57 Subtree Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
58 Random Walk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
59 One-Point List Crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
60 Two-Point List Crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
61 Duplicate Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
62 Simple Production Ruleset Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
63 Lexicographic Tournament Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
64 Double Tournament Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
65 Thread Pool Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
66 Fine-Grained Parallel Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
67 Simple Parallel Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
68 Simple Parallel Genetic Algorithm-style Breeding . . . . . . . . . . . . . . . . . . . . . . 96
69 Fine-Grained Parallel Genetic Algorithm-style Breeding . . . . . . . . . . . . . . . . . . . 96
70 An Abstract Generational Evolutionary Algorithm With Island Model Messaging . . . . 98
71 Fine-Grained Master-Side Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . 99
72 Threadsafe Collection Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
73 Asynchronous Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
74 Spatial Breeding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
75 Random Walk Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
76 An Abstract Generational 1-Population Competitive Coevolutionary Algorithm . . . . . 107
77 Pairwise Relative Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
78 Complete Relative Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
79 K-fold Relative Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
80 More Precise K-fold Relative Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . 109
81 Single-Elimination Tournament Relative Fitness Assessment . . . . . . . . . . . . . . . . 110
82 An Abstract Sequential 2-Population Competitive Coevolutionary Algorithm . . . . . . 112
83 K-fold Relative Fitness Assessment with an Alternative Population . . . . . . . . . . . . 113
84 An Abstract Parallel 2-Population Competitive Coevolutionary Algorithm . . . . . . . . 113
85 K-fold Relative Joint Fitness Assessment with an Alternative Population . . . . . . . . . 114
86 An Abstract Parallel Previous 2-Population Competitive Coevolutionary Algorithm . . 115
87 K-fold Relative Fitness Assessment with the Fittest of an Alternative Population . . . . . 115
88 An Abstract Sequential N-Population Cooperative Coevolutionary Algorithm (CCEA) . 118
5
89 K-fold Joint Fitness Assessment with N − 1 Collaborating Populations . . . . . . . . . . 118
90 An Abstract Parallel N-Population Cooperative Coevolutionary Algorithm . . . . . . . 119
91 K-fold Joint Fitness Assessment of N Populations . . . . . . . . . . . . . . . . . . . . . . 119
92 Implicit Fitness Sharing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
93 Deterministic Crowding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
94 Multiobjective Lexicographic Tournament Selection . . . . . . . . . . . . . . . . . . . . . 129
95 Multiobjective Ratio Tournament Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 130
96 Multiobjective Majority Tournament Selection . . . . . . . . . . . . . . . . . . . . . . . . 130
97 Multiple Tournament Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
98 Pareto Domination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
99 Pareto Domination Binary Tournament Selection . . . . . . . . . . . . . . . . . . . . . . . 132
100 Computing a Pareto Non-Dominated Front . . . . . . . . . . . . . . . . . . . . . . . . . . 132
101 Front Rank Assignment by Non-Dominated Sorting . . . . . . . . . . . . . . . . . . . . . 133
102 Multiobjective Sparsity Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
103 Non-Dominated Sorting Lexicographic Tournament Selection With Sparsity . . . . . . . 134
104 An Abstract Version of the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) . . 135
105 Compute the Distance of the Kth Closest Individual . . . . . . . . . . . . . . . . . . . . . 137
106 SPEA2 Archive Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
107 An Abstract Version of the Strength Pareto Evolutionary Algorithm 2 (SPEA2) . . . . . . 139
108 Greedy Randomized Adaptive Search Procedures (GRASP) . . . . . . . . . . . . . . . . . 145
109 An Abstract Ant Colony Optimization Algorithm (ACO) . . . . . . . . . . . . . . . . . . 146
110 The Ant System (AS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
111 Pheromone Updating with a Learning Rate . . . . . . . . . . . . . . . . . . . . . . . . . . 149
112 The Ant Colony System (ACS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
113 Guided Local Search (GLS) with Random Updates . . . . . . . . . . . . . . . . . . . . . . 154
114 An Abstract Version of the Learnable Evolution Model (LEM) . . . . . . . . . . . . . . . 156
115 Simple Rejection Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
116 Region-based Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
117 Weighted Rejection Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
118 An Abstract Estimation of Distribution Algorithm (EDA) . . . . . . . . . . . . . . . . . . 159
119 Population-Based Incremental Learning (PBIL) . . . . . . . . . . . . . . . . . . . . . . . . 162
120 The Compact Genetic Algorithm (cGA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
121 An Abstract Version of the Bayesian Optimization Algorithm (BOA) . . . . . . . . . . . 166
122 Q-Learning with a Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
123 Model-Free Q-Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
124 S AMUEL Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
125 Zeroth Classifier System Fitness Updating . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
126 Zeroth Classifier System Fitness Redistribution . . . . . . . . . . . . . . . . . . . . . . . . 187
127 The Zeroth Level Classifier System (ZCS) . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
128 XCS Fitness-Weighted Utility of an Action . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
129 XCS Best Action Determination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
130 XCS Action Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
131 XCS Fitness Updating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
132 XCS Fitness Redistribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
133 XCS Fitness Updating (Extended) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
6
0 Introduction
This is a set of lecture notes for an undergraduate class on metaheuristics. They were constructed
for a course I taught in Spring of 2009, and I wrote them because, well, there’s a lack of undergrad-
uate texts on the topic. As these are lecture notes for an undergraduate class on the topic, which is
unusual, these notes have certain traits. First, they’re informal and contain a number of my own
personal biases and misinformation. Second, they are light on theory and examples: they’re mostly
descriptions of algorithms and handwavy, intuitive explanations about why and where you’d want
to use them. Third, they’re chock full of algorithms great and small. I think these notes would best
serve as a complement to a textbook, but can also stand alone as rapid introduction to the field.
I make no guarantees whatsoever about the correctness of the algorithms or text in these notes.
Indeed, they’re likely to have a lot of errors. Please tell me of any errors you find (and correct!).
Some complex algorithms have been presented in simplified versions. In those cases I’ve noted it.
0.1 What is a Metaheuristic?

Metaheuristics is a rather unfortunate1 term often used to describe a major subfield, indeed
the primary subfield, of stochastic optimization. Stochastic optimization is the general class of
algorithms and techniques which employ some degree of randomness to find optimal (or as optimal
as possible) solutions to hard problems. Metaheuristics are the most general of these kinds of
algorithms, and are applied to a very wide range of problems.
What kinds of problems? In Jacobellis v. Ohio (1964, regarding obscenity), the United States
Supreme Court Justice Potter Stewart famously wrote,
I shall not today attempt further to define the kinds of material I understand to be embraced
within that shorthand description; and perhaps I could never succeed in intelligibly doing so.
But I know it when I see it, and the motion picture involved in this case is not that.
Metaheuristics are applied to I know it when I see it problems. They’re algorithms used to find
answers to problems when you have very little to help you: you don’t know what the optimal
solution looks like, you don’t know how to go about finding it in a principled way, you have very
little heuristic information to go on, and brute-force search is out of the question because the space
is too large. But if you’re given a candidate solution to your problem, you can test it and assess how
good it is. That is, you know a good one when you see it.
For example: imagine if you’re trying to find an optimal set of robot behaviors for a soccer
goalie robot. You have a simulator for the robot and can test any given robot behavior set and
assign it a quality (you know a good one when you see it). And you’ve come up with a definition
for what robot behavior sets look like in general. But you have no idea what the optimal behavior
set is, nor even how to go about finding it.
1 OrdinarilyI’d call the subfield stochastic optimization. But that’s too general a term; it includes important algorithms
like Markov Chain Monte Carlo (MCMC) or Gibbs Sampling, which are not in this category. Metaheuristics has lately
been the term of use, but I think it’s profoundly misleading and weird. When I hear “metadiscussion” I think: a discussion
about discussions. Likewise when I hear “metaheuristic” I think: a heuristic about (or for) heuristics. That’s not at all what
these algorithms are about! Perhaps the lesser-used term black box optimization would be better, though it too comes
with some additional baggage. Weak methods is also too broad a term: it doesn’t imply stochasticity. Sometimes the
term stochastic search is used: but I usually define “search” problems as all-or-nothing: either you find the solution or
you don’t. We’re not doing search; we’re doing optimization.
7
The simplest thing you could do in this situation is Random Search: just try random behavior
sets as long as you have time, and return the best one you discovered. But before you give up and
start doing random search, consider the following alternative, known as Hill-Climbing. Start with
a random behavior set. Then make a small, random modification to it and try the new version. If
the new version is better, throw the old one away. Else throw the new version away. Now make
another small, random modification to your current version (which ever one you didn’t throw
away). If this newest version is better, throw away your current version, else throw away the
newest version. Repeat as long as you can.
Hill-climbing is a simple metaheuristic algorithm. It exploits a heuristic belief about your space
of candidate solutions which is usually true for many problems: that similar solutions tend to
behave similarly (and tend to have similar quality), so small modifications will generally result
in small, well-behaved changes in quality, allowing us to “climb the hill” of quality up to good
solutions. This heuristic belief is one of the central defining features of metaheuristics: indeed,
nearly all metaheuristics are essentially elaborate combinations of hill-climbing and random search.
0.2 Algorithms
The lecture notes have a lot of algorithms, great and small. Everything from large evolutionary
computation algorithms to things as simple as “how to shuffle an array”. Algorithms appear for
even the most trivial and obvious of tasks. I strove to be pedantic in case anyone had any questions.
If the algorithm takes parameters, they will appear first followed by a blank line. If there are
no parameters, the algorithm begins immediately. In some cases the algorithm is actually several
functions, each labelled procedure. Sometimes certain shared, static global variables are defined
which appear at the beginning and are labelled global. Here is an example of a simple algorithm:
Algorithm 0 Bubble Sort

v ← $v1 , ..., vl % vector to sort
1: ! " User-provided parameters to the algorithm appear here
" Then a blank space
2: repeat " Algorithm begins here
3: swapped ← false " ← always means “is set to”
4: for i from 1 to l − 1 do " Note that l is defined by vl in Line 1
5: if vi > vi+1 then
6: Swap vi and vi+1
7: swapped ← true
8: until swapped = false " = means “is equal to”
9: return !v " Some algorithms return nothing, so there is no return statement
Note that the parameters to the function are only loosely specified: and sometimes when we
call a function, we don’t explicitly state the parameters, if it’s obvious what needs to be provided.
Yeah, I could have been more formal in a lot of places. So sue me.2
2 Don’t sue me. Thanks.
8
0.3 Notation3
There’s little special here. But just to dot our i’s and cross our t’s:
• Numbers and booleans are denoted with lower-case letters, greek symbols, or words (n, λ,
min, popsize). The default “empty” or “null” element is denoted with !. Ranges of numbers
are often described like this: from 1 to n inclusive. Ranges can be of integers or real values.
The symbol ← always means “is set to”, and the symbol = usually means “equals”.
• Candidate solutions (sometimes called individuals, particles, or trails) are indicated with
upper-case letters or words (Best, S, Pi ). Some candidate solutions are actually vectors and are
described like vectors below. Others consist of a number of components, often designated
C1 ...Cj . Candidate solutions may be associated with some kind of quality (fitness), usually
via a function like Quality(S) or Fitness(Pi ). Quality can be set as well. Usually quality is a
single number; but can in some cases (for multiobjective optimization) be a group of numbers
called objectives. The value of objective O j , assigned to individual Pi , is accessed via a
function like ObjectiveValue(O j , Pi ). In certain cases various other attributes may be assigned
to individuals or to other objects.
• Collections (or bags, or groups, or pools, or lists, or multisets) are groups of objects where
the objects usually don’t have to be unique. In fact, the lecture notes rarely use sets, and
abundantly use collections. A collection is denoted with a capital letter like P and contains
some number of elements in braces { P1 , ..., Pn }. The size of P is || P|| or (in this case) n.
Membership (or lack thereof) is indicated with ∈ or ∈ /. Usually there is an implicit order in a
collection, so you can refer to its elements uniquely (P4 or Pi ) and can scan through it like this
(for each element Pi ∈ P do ... Pi ) or this (for i from 1 to n do ... Pi ). Collections are generally
read-only, though their elements may allow internal modification.
The union operator (∪) is abused to indicate concatenating collections (like P ← P ∪ Q).
This is often used to add an element to a collection like P ← P ∪ { R j }. The minus sign
is abused to indicate removing all elements that are in another collection, as in P − M, or
removing a specific element from a collection (P ← P − { P2 }). In all these cases, presume
that the new collection retains the implicit order from the old collection or collections.
The most common collections are the ones used for populations. Usually I denote popu-
lations with P or Q. Occasionally we need to have a collection of populations, denoted like
(i )
this: P(1) , ..., P(n) . An individual number j in population P(i) would be Pj .
Sometimes children will be denoted Ca and Cb , etc. This doesn’t imply the existence of a
collection called C (though it’s generally harmless to do so).
• First-in, First-out Queues are treated like collections, with the additional ability to add to the
end of them and remove elements from the front or from an arbitrary location.
−→
• Vectors are denoted with an over-arrow (!x, Best) and contain some number of elements in
angle brackets $ x1 , ..., xn %. Unlike collections, vectors are modifiable. An element in a vector
can be replaced with another object at the same location. Slots may not be simply deleted
from a vector, but vectors can be extended by adding elements to the end of them. I use
vectors instead of collections when we must explicitly change elements in certain locations.
• Tuples are vectors with named slots like !t ← $tlock , tdata %, rather than numbered slots.
3 This is always the most boring part of a book! Why are you reading this?
9
• Two-dimensional Arrays or Matrices are denoted with capital letters (A) and their elements
can be referred to in the usual manner: Ai,j . Like vectors, array elements can be replaced.
• Probability Distributions and other Models are denoted with a capital letter like T. Distri-
butions and Models are constructed or updated; then we select random numbers under them.
Along those lines, variances are denoted with σ2 , standard deviations with σ, and means are
often denoted with µ.
• When passed around as data, Functions are in lower-case, as in f or f (node).
10
1 Gradient-based Optimization
Before we get into metaheuristics, let’s start with a traditional
mathematical method for finding the maximum of a function: f( )
Gradient Ascent.4 The idea is to identify the slope and move up
it. Simple! The function we’re going to maximize is f ( x ). This new location
method doesn’t require us to compute or even know f ( x ), but it
does assume we can compute the slope of x, that is, we have f ) ( x ). old location
The technique is very simple. We start with an arbitrary value slope f'(x) !x,f(x)〉
for x. We then repeatedly add to it a small portion of its slope, that
is, x ← x + α f ) ( x ), where α is a very small positive value. If the
slope is positive, x will increase. If the slope is negative, x will
decrease. Figure 1 roughly illustrates this. Ultimately x will move Figure 1 Gradient Ascent with a
negative slope. x is decreasing.
up the function until it is at the peak, at which point the slope is
zero and x won’t change any more.
We’re usually not interested in simple one-dimensional functions like this: more generally, we’d
like to find the maximum of a multidimensional function. To do this we replace x with the vector !x,
and replace the slope f ) ( x ) with the gradient of !x, ∇ f (!x ). As a reminder: the gradient is simply a
∂f ∂f ∂f
vector where each element is the slope of !x in that dimension, that is, $ ∂x1 , ∂x2 , ..., ∂xn %. So basically
we’re going up the slope in all dimensions at once. Here’s the Gradient Ascent algorithm in its full
five-line glory:
Algorithm 1 Gradient Ascent

x ← random initial vector
1: !
2: repeat
3: !x ← !x + α∇ f (!x ) " In one dimension: x ← x + α f ) ( x )
x is the ideal solution or we have run out of time
4: until !
5: return !x
Note that the algorithm runs until we’ve found “the ideal solu- f( )
tion” or “we have run out of time”. How do we know that we’ve
got the ideal solution? Typically when the slope is 0. However saddle point
there are points besides maxima where this is the case: the minima
of functions (of course) and also saddle points such as in Figure 2.
One issue with Gradient Ascent is convergence time. As we
get close to the maximum of the function, Gradient Ascent will
overshoot the top and land on the other side of the hill. It may
overshoot the top many times, bouncing back and forth as it moves
closer to the maximum. Figure 3 shows this situation.
One of the reasons for this is that the size of the jumps Gradient Figure 2 A saddle point.
Ascent makes is entirely based on the current slope. If the slope is
very steep the jump will be large even if it’s not warranted. One way to deal with this is to tune
4 Actually, the method is usually called Gradient Descent because it’s used to find the minimum of a function. To do
that, we just subtract the gradient or slope rather than add it, that is, Algorithm 1 has its line changed to !x ← !x − α∇ f (!x ).
But in our later examples we’re always finding maxima, so we’re going to be consistent here.
11
Gradient Ascent for your problem, by adjusting the value of α. A very small value of α and Gradient
Ascent won’t overshoot hills but it may take a long time to march up the hills and converge to the
top. But a very big value of α will cause Gradient Ascent to constantly overshoot the hills which
also causes it to take a long time to converge to the maximum, if at all. We’re looking for a value of
α which is “just right”.
We could also modify the algorithm to consider other fac-
tors. For example, if we could compute not only f ) ( x ) but f( )
also f )) ( x ), we could use Newton’s Method.5 This varia-
tion on Gradient Ascent includes an additional f ))1( x) like so:
f ) (x)
x ← x + α f )) ( x) . This modification dampens α as we approach
a zero slope.
The multidimensional situation is not so simple. The mul-
tidimensional version of a first derivative f ) ( x ) is the gradient
∇ f (!x ). But the multidimensional version of a second derivative
f )) ( x ) is a complex matrix called a Hessian H f (!x ) consisting of
partial second derivatives along each dimension. The Hessian
is shown in Figure 4. new x old x
To make matters worse, we’re dividing by the second deriva- Figure 3 Gradient Ascent overshooting
tive, which in the multidimensional case involves finding the the maximum.
inverse of this matrix. Overall, the method looks like this:
Algorithm 2 Newton’s Method (Adapted for Maximum Finding)

x ← random initial vector
1: !
2: repeat
)
3: !x ← !x + α[ H f (!x )]−1 ∇ f (!x ) " In one dimension: x ← x + α ff))((xx))
x is the ideal solution or we have run out of time
4: until !
5: return !x
Because it employs the second derivative, Newton’s

Method generally converges faster than regular Gra-  
∂ ∂f ∂ ∂f ∂ ∂f
dient Ascent, and it also gives us the information to  ∂x1 ∂x1 ∂x1 ∂x2 · · · ∂x1 ∂xn 
 
determine if we’re at the top of a local maximum (as  
 ∂ ∂ f ∂ ∂ f ∂ ∂ f 
opposed to a minimum or saddle point) because at a  ∂x2 ∂x1 ∂x2 ∂x2 · · · ∂x2 ∂xn 

H f (!x ) =  
maximum, f ) ( x ) is zero and f )) ( x ) is negative.  .. .. .. ..  
But even so, this doesn’t get around the real problem  . . . . 
 
with these methods: they get caught in local optima.  
∂ ∂f ∂ ∂f ∂ ∂f
Local optima of a function are the optima (in our case, ∂xn ∂x1 ∂xn ∂x2 · · · ∂xn ∂xn
maxima) of a local region. Global optima are the op-
tima of the entire function. Figure 5 shows the trace
Figure 4 The Hessian H f (!x ).
of Gradient Ascent getting caught in a local optimum.
Gradient Ascent and Newton’s Method are local opti-
mization algorithms.
5 Asin Sir Isaac Newton, 1642–1727. This method is normally used to find the zeros (roots) of functions, but it’s easily
modified to find maxima, as we’ve done here.
12
How do you escape local optima?
f( )
With the tools we have so far, there’s
really only one way: change α to a suf-
ficiently large value that the algorithm
potentially overshoots not only the top
of its hill but actually lands on the next I'm at the top of
the world!
hill.
Alternatively, we could put New-
ton’s Method in a big loop trying over
and over again with different starting
points, and finally returning the best so-
lution discovered. To determine what
the “best solution discovered” is, we
need to be able to compute f ( x ) (some-
thing we’ve not required up till now) so
we can compare results. Assuming we
Figure 5 Gradient Ascent stuck in a local optimum.
have that, we can now construct a global
optimization algorithm.
Algorithm 3 Newton’s Method with Restarts (One Dimensional Version)

1: x ← random initial value
2: x ∗ ← x " x ∗ will hold our best discovery so far.
3: repeat
4: repeat
f ) (x)
5: x ← x + α f )) ( x)
6: until f ) ( x ) = 0 and f )) ( x ) < 0 " That is, until we’re at the maximum.
7: if f ( x ) > f ( x ∗ ) then " Found a new best result!
8: x∗ ← x
9: x ← random value
10: until we have run out of time
11: return x ∗
A global optimization algorithm is guaranteed to find the global optimum if it runs long enough.
The above algorithm is really global only in theory: we’ll likely never have f ) ( x ) precisely equal to 0.
So we’ll have to fudge it: if −% < f ) ( x ) < % for some very small value of %, we’ll consider that
“close enough to zero”.6
I’ll leave it as an exercise to the reader to figure out what the multidimensional version of this
algorithm is.
6 Thereis a gotcha with the algorithms described here: what happens when part of the function is totally flat? There’s
no gradient to ascend, which leads to some problems. Let’s say you’re in a flat valley (a local minimum) of the function.
All around you the slope is 0, so Gradient Ascent won’t move at all. It’s stuck. Even worse: the second derivative is
f ) (x)
0 as well, so for Newton’s Method, f )) ( x) = 00 . Eesh. And to top it off, f ”( x ) = f )) ( x ) = 0 for flat minima, flat saddle
points, and flat maxima. Perhaps adding a bit of randomness might help in some of these situations: but that’s for the
next section....
13
14
2 Single-State Methods
Gradient-based optimization makes a big assumption: that you can compute the first (or even the
second) derivative. That’s a big assumption. If you are optimizing a well-formed, well-understood
mathematical function, it’s reasonable. But in most cases, you can’t compute the gradient of the
function because you don’t even know what the function is. All you have is a way of creating or
modifying inputs to the function, testing them, and assessing their quality.
For example, imagine that you have a humanoid robot simulator, and you’re trying to find
an optimal loop of timed operations to keep the robot walking forward without falling over. You
have some n different operations, and your candidate solutions are arbitrary-length strings of these
operations. You can plug a string in the simulator and get a quality out (how far the robot moved
forward before it fell over). How do you find a good solution?
All you’re given is a black box (in this case, the robot simulator) describing a problem that you’d
like to optimize. The box has a slot where you can submit a candidate solution to the problem
(here, a string of timed robot operations). Then you press the big red button and out comes the
assessed quality of that candidate solution. You have no idea what kind of surface the quality
assessment function looks like when plotted. Your candidate solution doesn’t even have to be a
vector of numbers: it could be a graph structure, or a tree, or a set of rules, or a string of robot
operations! Whatever is appropriate for the problem.
To optimize a candidate solution in this scenario, you need to be able to do four things:
• Provide one or more initial candidate solutions. This is known as the initialization procedure.
• Assess the Quality of a candidate solution. This is known as the assessment procedure.
• Make a Copy of a candidate solution.
• Tweak a candidate solution, which produces a randomly slightly different candidate solution.
This, plus the Copy operation are collectively known as the modification procedure.
To this the metaheuristics algorithm will typically provide a selection procedure that decides
which candidate solutions to retain and which to reject as it wanders through the space of possible
solutions to the problem.
2.1 Hill-Climbing
Let’s begin with a simple technique, Hill-Climbing. This technique is related to gradient ascent,
but it doesn’t require you to know the strength of the gradient or even its direction: you just
iteratively test new candidate solutions in the region of your current candidate, and adopt the new
ones if they’re better. This enables you to climb up the hill until you reach local optima.
Algorithm 4 Hill-Climbing
1: S ← some initial candidate solution " The Initialization Procedure
2: repeat
3: R ← Tweak(Copy(S)) " The Modification Procedure
4: if Quality(R) > Quality(S) then " The Assessment and Selection Procedures
5: S←R
6: until S is the ideal solution or we have run out of time
7: return S
15
Notice the strong resemblance between Hill-Climbing and Gradient Ascent. The only real
difference is that Hill-Climbing’s more general Tweak operation must instead rely on a stochastic
(partially random) approach to hunting around for better candidate solutions. Sometimes it finds
worse ones nearby, sometimes it finds better ones.
We can make this algorithm a little more aggressive: create n “tweaks” to a candidate solution
all at one time, and then adopt the best one. This modified algorithm is called Steepest Ascent
Hill-Climbing, because by sampling all around the original candidate solution and then picking
the best, we’re essentially sampling the gradient and marching straight up it.
Algorithm 5 Steepest Ascent Hill-Climbing
1: n ← number of tweaks desired to sample the gradient
2: S ← some initial candidate solution

3: repeat
4: R ← Tweak(Copy(S))
5: for n − 1 times do
6: W ← Tweak(Copy(S))
7: if Quality(W) > Quality(R) then
8: R←W
9: if Quality(R) > Quality(S) then
10: S←R
11: until S is the ideal solution or we have run out of time
12: return S
A popular variation, which I dub Steepest Ascent Hill-Climbing with Replacement, is to not
bother comparing R to S: instead, we just replace S directly with R. Of course, this runs the risk of
losing our best solution of the run, so we’ll augment the algorithm to keep the best-discovered-so-
far solution stashed away, in a reserve variable called Best. At the end of the run, we return Best. In
nearly all future algorithms we’ll use the store-in-Best theme, so get used to seeing it!
Algorithm 6 Steepest Ascent Hill-Climbing With Replacement

3: Best ← S
4: repeat
9: R←W
10: S←R
11: if Quality(S) > Quality(Best) then
12: Best ← S
13: until Best is the ideal solution or we have run out of time
14: return Best
16
2.1.1 The Meaning of Tweak
The initialization, Copy, Tweak, and (to a lesser extent) fitness assessment functions collectively
define the representation of your candidate solution. Together they stipulate what your candidate
solution is made up of and how it operates.
What might a candidate solution look like? It could be a vector; or an arbitrary-length list of
objects; or an unordered set or collection of objects; or a tree; or a graph. Or any combination of
these. Whatever seems to be appropriate to your problem. If you can create the four functions
above in a reasonable fashion, you’re in business.
One simple and common representation for candidate solutions, which we’ll stick to for now, is
the same as the one used in the gradient methods: a fixed-length vector of real-valued numbers.
Creating a random such vector is easy: just pick random numbers within your chosen bounds. If
the bounds are min and max inclusive, and the vector length is l, we could do this:
Algorithm 7 Generate a Random Real-Valued Vector

1: min ← minimum desired vector element value
2: max ← maximum desired vector element value
3: !v ← a new vector $v1 , v2 , ...vl %

4: for i from 1 to l do
5: vi ← random number chosen uniformly between min and max inclusive
6: return !v
To Tweak a vector we might (as one of many possibilities) add a small amount of random
noise to each number: in keeping with our present definition of Tweak, let’s assume for now that
this noise is no larger than a small value. Here’s a simple way of adding bounded, uniformly
distributed random noise to a vector. For each slot in the vector, if a coin-flip of probability p comes
up heads, we find some bounded uniform random noise to add to the number in that slot. In most
cases we keep p = 1.
Algorithm 8 Bounded Uniform Convolution

v ← vector $v1 , v2 , ...vl % to be convolved
1: !
2: p ← probability of adding noise to an element in the vector " Often p = 1
3: r ← half-range of uniform noise
7: if p ≥ random number chosen uniformly from 0.0 to 1.0 then
8: repeat
9: n ← random number chosen uniformly from −r to r inclusive
10: until min ≤ vi + n ≤ max
11: vi ← vi + n
12: return !v
17
Unimodal Needle in a Haystack
Noisy Deceptive
(or “Hilly” or “Rocky”)
Figure 6 Four example quality functions.
We now have a knob we can turn: r, the size of the bound on Tweak. If the size is very small,
then Hill-Climbing will march right up a local hill and be unable to make the jump to the next hill
because the bound is too small for it to jump that far. Once it’s on the top of a hill, everywhere it
jumps will be worse than where it is presently, so it stays put. Further, the rate at which it climbs
the hill will be bounded by its small size. On the other hand, if the size is large, then Hill-Climbing
will bounce around a lot. Importantly, when it is near the top of a hill, it will have a difficult time
converging to the peak, as most of its moves will be so large as to overshoot the peak.
Thus small sizes of the bound move slowly and get caught in local optima; and large sizes on
the bound bounce around too frenetically and cannot converge rapidly to finesse the very top of
peaks. Notice how similar this is to α used in Gradient Ascent. This knob is one way of controlling
the degree of Exploration versus Exploitation in our Hill-Climber. Optimization algorithms which
make largely local improvements are exploiting the local gradient, and algorithms which mostly
wander about randomly are thought to explore the space. As a rule of thumb: you’d like to use a
highly exploitative algorithm (it’s fastest), but the “uglier” the space, the more you will have no
choice but to use a more explorative algorithm.
2.2 Single-State Global Optimization Algorithms

A global optimization algorithm is one which, if we run it long enough, will eventually find the
global optimum. Almost always, the way this is done is by guaranteeing that, at the limit, every
location in the search space will be visited. The single-state algorithms we’ve seen so far cannot
guarantee this. This is because of our definition (for the moment) of Tweak: to “make a small,
bounded, but random change”. Tweak wouldn’t ever make big changes. If we’re stuck in a
sufficiently broad local optimum, Tweak may not be strong enough to get us out of it. Thus the
algorithms so far have been local optimization algorithms.
18
There are many ways to construct a global optimization algorithm instead. Let’s start with the
simplest one possible: Random Search.
Algorithm 9 Random Search

1: Best ← some initial random candidate solution
2: repeat
3: S ← a random candidate solution
5: Best ← S
7: return Best
Random Search is the extreme in exploration (and global optimization); in contrast, Hill-
Climbing (Algorithm 4), with Tweak set to just make very small changes and never make large ones,
may be viewed as the extreme in exploitation (and local optimization). But there are ways to achieve
reasonable exploitation and still have a global algorithm. Consider the following popular technique,
called Hill-Climbing with Random Restarts, half-way between the two. We do Hill-Climbing for
a certain random amount of time. Then when time is up, we start over with a new random location
and do Hill-Climbing again for a different random amount of time. And so on. The algorithm:
Algorithm 10 Hill-Climbing with Random Restarts

1: T ← distribution of possible time intervals
2: S ← some initial random candidate solution

3: Best ← S
4: repeat
5: time ← random time in the near future, chosen from T
6: repeat
9: S←R
10: until S is the ideal solution, or time is up, or we have run out of total time
12: Best ← S
13: S ← some random candidate solution
14: until Best is the ideal solution or we have run out of total time
15: return Best
If the randomly-chosen time intervals are generally extremely long, this algorithm is basically
one big Hill-Climber. Likewise, if the intervals are very short, we’re basically doing random search
(by resetting to random new locations each time). Moderate interval lengths run the gamut between
the two. That’s good, right?
It depends. Consider Figure 6. The first figure, labeled Unimodal, is a situation where Hill-
Climbing is close to optimal, and where Random Search is a very bad pick. But for the figure
labelled Noisy, Hill-Climbing is quite bad; and in fact Random Search is expected to be about
as good as you can do (not knowing anything about the functions beforehand). The difference
19
is that in Unimodal there is a strong relationship between the distance (along the x axis) of two
candidate solutions and their relationship in quality: similar solutions are generally similar in
quality, and dissimilar solutions don’t have any relationship per se. In the Noisy situation, there’s
no relationship like this: even similar solutions are very dissimilar in quality. This is often known
as the smoothness criterion for local search to be effective.
This isn’t sufficient though. Consider the figure labeled Needle in a Haystack, for which Random
Search is the only real way to go, and Hill-Climbing is quite poor. What’s the difference between
this and Unimodal? After all, Needle in a Haystack is pretty smooth. For local search to be effective
there must be an informative gradient which generally leads towards the best solutions. In fact,
you can make highly uninformative gradients for which Hill-Climbing is spectacularly bad! In the
figure labeled Deceptive, Hill-Climbing not only will not easily find the optimum, but it is actively
let away from the optimum.
Thus there are some kinds of problems where making small local greedy changes does best; and
other problems where making large, almost random changes does best. Global search algorithms
run this gamut, and we’ve seen it before: Exploration versus Exploitation. Once again, as a rule of
thumb: you’d like to use a highly exploitative algorithm (it’s fastest), but the “uglier” the space, the
more you will have no choice but to use a more explorative algorithm.
Here are some ways to create a global search algorithm, plus approaches to tweaking exploration
vs. exploitation within that algorithm:
• Adjust the Modification procedure Tweak occasionally makes large, random changes.
Why this is Global If you run the algorithm long enough, this randomness will cause Tweak
to eventually try every possible solution.
Exploration vs. Exploitation The more large, random changes, the more exploration.
• Adjust the Selection procedure Change the algorithm so that you can go down hills at
least some of the time.
Why this is Global If you run the algorithm long enough, you’ll go down enough hills that
you’ll eventually find the right hill to go up.
Exploration vs. Exploitation The more often you go down hills, the more exploration.
• Jump to Something New Every once in a while start from a new location.
Why this is Global If you try enough new locations, eventually you’ll hit a hill which has the
highest peak.
Exploration vs. Exploitation The more frequently you restart, the more exploration.
• Use a Large Sample Try many candidate solutions in parallel.

Why this is Global With enough parallel candidate solutions, one of them is bound to be on
the highest peak.
Exploration vs. Exploitation More parallel candidate solutions, more exploration.
Let’s look at some additional global optimizers. We’ll focus on what I’m calling single-state
optimizers which only keep around one candidate solution at a time. That is: no large sample.
20
2.3 Adjusting the Modification Procedure: (1+1), (1+λ), and (1, λ)
These three oddly named algorithms are forms of our Hill-Climbing procedures with variations
of the Tweak operation to guarantee global optimization. They’re actually degenerate cases of the
more general (µ, λ) and (µ + λ) evolutionary algorithms discussed later (in Section 3.1).
The goal is simple: construct a Tweak operation which tends to tweak in small ways but
occasionally makes larger changes, and can make any possible change. We’ll mostly hill-climb, but
also have the ability to, occasionally, jump far enough to land on other peaks. And there is a chance,
however small, that the Hill-Climber will get lucky and Tweak will land right on the optimum.
For example, imagine that we’re back to represent-
ing solutions in the form of fixed-length vectors of real
5
numbers. Previously our approach to Tweaking vec-
tors was Bounded Uniform Convolution (Algorithm 8). 4
The key word is bounded: it required you to choose be- 3
tween being small enough to finesse local peaks and
being large enough to escape local optima. But a Gaus- 2
sian7 (or Normal, or bell curve) distribution N (µ, σ2 ) 1
lets you do both: usually it makes small numbers but
sometimes it makes large numbers. Unless bounded, a 0 -1 -0.5 0 0.5 1
Gaussian distribution will occasionally make very large Figure 7 Three Normal or Gaussian distributions
2
numbers indeed. The distribution requires two parame- N2(µ, σ ) with the mean µ = 0 and the variance
2 σ set to σ2 = 0.005: ——, σ2 = 0.02: – – –, and
ters: the mean µ (usually 0) and variance σ . The degree σ2 = 0.1: - - - -.
to which we emphasize small numbers over large ones
can be controlled by simply changing the variance σ2 of the distribution.
We can do this by adding to each number in the vector some random noise under a Gaussian
distribution with a mean µ = 0. This is called Gaussian convolution. Most noise will be close to 0,
so the vector values won’t change much. But occasional values could be quite large. As was the
case for Algorithm 8, each slot receives noise only with a probability p.
Algorithm 11 Gaussian Convolution
1: !
2: p ← probability of adding noise to an element in the vector " Often p = 1
3: σ2 ← variance of Normal distribution to convolve with " Normal = Gaussian
8: repeat
9: n ← random number chosen from the Normal distribution N (0, σ2 )
11: vi ← vi + n
12: return !v
7 Karl Friedrich Gauss, 1777–1855, kid genius, physicist, and possibly the single most important mathematician ever.
21
(1+1) is the name we give to standard Hill-Climbing (Algorithm 4) with this probabilistic-
modified Tweak. (1+λ) is the name we give to a similarly modified Steepest Ascent Hill-Climbing
(Algorithm 5). And (1, λ) is the name we give to the modified Steepest Ascent Hill-Climbing with
Replacement (Algorithm 6). These names seem cryptic now but will make more sense later.
As it turns out, Gaussian Convolution doesn’t give
us just one new knob (σ2 ) to adjust exploration vs. ex- Noise in Tweak
ploitation, but two knobs. Consider the Steepest Ascent Low High
Hill-Climbing with Replacement algorithm (Algorithm 6),
Samples
where the value n specified how many children are gen- Few Explorative
.
erated from the parent candidate solution through Tweak.
Many Exploitative
In the “global” version of this algorithm, (1, λ), the value
of n interacts with σ2 in an important way: if σ2 is large Table 1 Simplistic description of the interaction
(noisy), then the algorithm will search crazier locations: of two factors and their effect on exploration
versus exploitation. The factors are: degree of
but a high value of n will aggressively weed out the poor
noise in the Tweak operation; and the samples
candidates discovered at those locations. This is because taken before adopting a new candidate solution.
if n is low, a poor quality candidate may still be the best
of the n examined; but if n is high, this is much less likely. Thus while σ2 is pushing for more
exploration (at the extreme: random search), a high value of n is pushing for more exploitation. n
is an example of what will later be called selection pressure. Table 1 summarizes this interaction.
Many random number generators provide facilities for selecting random numbers under
Normal (Gaussian) distributions. But if yours doesn’t, you can make two Gaussian random
numbers at a time using the Box-Muller-Marsaglia Polar Method.8
Algorithm 12 Sample from the Gaussian Distribution (Box-Muller-Marsaglia Polar Method)

1: µ ← desired mean of the Normal distribution " Normal = Gaussian
2
2: σ ← desired variance of the Normal distribution
3: repeat
4: x ← random number chosen uniformly from 0.0 to 1.0
5: y ← random number chosen uniformly from 0.0 to 1.0 " x and y should be independent
6: 2
w ← x +y 2
7: until 0 < w '

<1 " Else we could divide by zero or take the square root of a negative number!
ln w
√
8: g ← µ + xσ −2 w " It’s σ, that is, σ2 . Also, note that ln is loge
'
9: h ← µ + yσ −2 lnww " Likewise.
10: return g and h " This method generates two random numbers at once. If you like, just use one.
Some random number generators (such as java.util.Random) only provide Gaussian random
numbers from the standard normal distribution N (0, 1). You can convert these numbers to a
Gaussian distribution for any mean µ and variance σ2 or standard deviation σ you like very simply:
√
N (µ, σ2 ) = µ + σ2 N (0, 1) = µ + σN (0, 1)
8 The method was first described in George Edward Pelham Box and Mervin Muller, 1958, A note on the generation of
random normal deviates, The Annals of Mathematical Statistics, 29(2), 610–611. However the polar form of the method, as
shown here, is usually ascribed to the mathematician George Marsaglia. There is a faster, but not simpler, method with a
great, and apt, name: the Ziggurat Method.
22
2.4 Simulated Annealing
Simulated Annealing was developed by various researchers in the mid 1980s, but it has a famous
lineage, being derived from the Metropolis Algorithm, developed by the ex-Manhattan Project
scientists Nicholas Metropolis, Arianna and Marshall Rosenbluth, and Augusta and Edward Teller
in 1953.9 The algorithm varies from Hill-Climbing (Algorithm 4) in its decision of when to replace
S, the original candidate solution, with R, its newly tweaked child. Specifically: if R is better than
S, we’ll always replace S with R as usual. But if R is worse than S, we may still replace S with R
with a certain probability P(t, R, S):
Quality(R)−Quality(S)
P(t, R, S) = e t
where t ≥ 0. That is, the algorithm sometimes goes down hills. This equation is interesting in two
ways. Note that the fraction is negative because R is worse than S. First, if R is much worse than S,
the fraction is larger, and so the probability is close to 0. If R is very close to S, the probability is
close to 1. Thus if R isn’t much worse than S, we’ll still select R with a reasonable probability.
Second, we have a tunable parameter t. If t is close to 0, the fraction is again a large number,
and so the probability is close to 0. If t is high, the probability is close to 1. The idea is to initially set
t to a high number, which causes the algorithm to move to every newly-created solution regardless
of how good it is. We’re doing a random walk in the space. Then t decreases slowly, eventually to
0, at which point the algorithm is doing nothing more than plain Hill-Climbing.
Algorithm 13 Simulated Annealing
1: t ← temperature, initially a high number

3: Best ← S
4: repeat
Quality(R)−Quality(S)
6: if Quality(R) > Quality(S) or if a random number chosen from 0 to 1 < e t then
7: S←R
8: Decrease t
10: Best ← S
11: until Best is the ideal solution, we have run out of time, or t ≤ 0
12: return Best
The rate at which we decrease t is called the algorithm’s schedule. The longer we stretch out
the schedule, the longer the algorithm resembles a random walk and the more exploration it does.
9 Nicholas Metropolis, Arianna Rosenbluth, Marshall Rosenbluth, Augusta Teller, and Edward Teller, 1953, Equation
of state calculations by fast computing machines, Journal of Chemical Physics, 21, 1087–1091. And yes, Arianna and
Marshall were married, as were Augusta and Edward. Now that’s a paper! This gang also developed the Monte Carlo
Method widely used in simulation. Edward Teller later became a major advocate for nuclear testing and is believed to
be one of the inspirations for Dr. Strangelove. To make this Gordion knot even more convoluted, Augusta and Edward’s
grandson Eric Teller, who goes by Astro Teller, did a fair bit of early work in Genetic Programming! (See Section 4.3). A
later paper on Simulated Annealing which established it as a real optimization algorithm is Scott Kirkpatrick, Charles
Daniel Gelatt Jr., and Mario Vecchi, 1983, Optimization by simulated annealing, Science, 220(4598), 671–680.
23
Simulated Annealing gets its name from annealing, a process of cooling molten metal. If you
let metal cool rapidly, its atoms aren’t given a chance to settle into a tight lattice and are frozen in a
random configuration, resulting in brittle metal. If we decrease the temperature very slowly, the
atoms are given enough time to settle into a strong crystal. Not surprisingly, t means temperature.
2.5 Tabu Search

Tabu Search, by Fred Glover,10 employs a different approach to doing exploration: it keeps around
a history of recently considered candidate solutions (known as the tabu list) and refuses to return to
those candidate solutions until they’re sufficiently far in the past. Thus if we wander up a hill, we
have no choice but to wander back down the other side because we’re not permitted to stay at or
return to the top of the hill.
The simplest approach to Tabu Search is to maintain a tabu list L, of some maximum length l,
of candidate solutions we’ve seen so far. Whenever we adopt a new candidate solution, it goes
in the tabu list. If the tabu list is too large, we remove the oldest candidate solution and it’s no
longer taboo to reconsider. Tabu Search is usually implemented as a variation on Steepest Ascent
with Replacement (Algorithm 6). In the version below, we generate n tweaked children, but only
consider the ones which aren’t presently taboo. This requires a few little subtle checks:
Algorithm 14 Tabu Search

1: l ← Desired maximum tabu list length

4: Best ← S
5: L ← {} a tabu list of maximum length l " Implemented as first in, first-out queue
6: Enqueue S into L
7: repeat
8: if Length(L) > l then
9: Remove oldest element from L
13: if W ∈ / L and (Quality(W) > Quality(R) or R ∈ L) then
14: R←W
15: if R ∈/ L and Quality(R) > Quality(S) then
16: S←R
17: Enqueue R into L
19: Best ← S
21: return Best
10 “Tabu” is an alternate spelling for “taboo”. Glover also coined the word “metaheuristics”, and developed Scatter
Search with Path Relinking (Section 3.3.5). Tabu Search showed up first in Fred Glover, 1986, Future paths for integer
programming and links to artificial intelligence, Computers and Operations Research, 5, 533–549.
24
Tabu Search really only works in discrete spaces. What if your search space is real-valued
numbers? Only in truly exceptional situations will you visit the same real-valued point in space
twice, making the tabu list worthless. In this situation, one approach is to consider a solution to
be a member of a list if it is “sufficiently similar” to an existing member of the list. The similarity
distance measure will be up to you. See Section 6.4 for some ideas.
Even so, the big problem with Tabu Search is that if your search space is very large, and
particularly if it’s of high dimensionality, it’s easy to stay around in the same neighborhood, indeed
on the same hill, even if you have a very large tabu list. There may be just too many locations. An
alternative approach is to create a tabu list not of candidate solutions you’ve considered before, but
of changes you’ve made recently to certain features. For example, imagine if you’re finding a solution
to a graph problem like the Traveling Salesman Problem (see Section 8). You tweak a candidate
solution to create a new one, by deleting edge A and adding edges B and C, and decide to adopt
the new solution. Instead of placing the solution into the tabu list, you place the changes you made
into the list. A, B, and C each go into the list. Now for a while, while you’re thinking about new
tweaks, you’re not allowed to even consider adding or deleting A, B, or C. They’re taboo for now.
To implement this, the big change we’ll need to make is in the nature of the queue acting as our
tabu list. No longer can the queue be a simple first-in first-out queue because variable numbers
of things will enter the queue at any time step. Instead we’ll implement it as a set of tuples $ X, d%
where X is a feature we changed (for example “Edge A”), and d is the timestamp of when we made
the change. Also, we can no longer simply test for membership in the queue. Instead, we’ll have to
hand the queue to the Tweak operation, so it knows which changes it’s not allowed to make. Thus
our revised version: Tweak(Copy(...), L). I call the new algorithm Feature-based Tabu Search.
Algorithm 15 Feature-based Tabu Search
1: l ← desired queue length

4: Best ← S
5: L ← {} " L will hold tuples of the form $ X, d% where X is a feature and d is a timestamp
6: c←0
7: repeat
8: c ← c+1
9: Remove from L all tuples of the form $ X, d% where c − d > l " The “old” ones
10: R ← Tweak(Copy(S), L) " Tweak will not shift to a feature in L
12: W ← Tweak(Copy(S), L)
14: R←W
15: S←R
16: for each feature X modified by Tweak to produce R from S do
17: L ← L ∪ {$ X, c%}
19: Best ← S
21: return Best
25
Feature-based Tabu Search is somewhat different from the other techniques described here in
that it relies on the identifiability and separability of features found in candidate solutions, rather
than considering each candidate solution as an atomic element except for Tweak purposes. We’ll
see this notion put to more heavy use in Combinatorial Optimization (Section 8).
2.6 Iterated Local Search

This is the present name for a concept which has been around, in many guises, since at least the
1980s.11 It’s essentially a more clever version of Hill-Climbing with Random Restarts. Each time
you do a random restart, the hill-climber then winds up in some (possibly new) local optimum.
Thus we can think of Hill-Climbing with Random Restarts as doing a sort of random search through
the space of local optima. We find a random local optimum, then another, then another, and so on,
and eventually return the best optimum we ever discovered (ideally, it’s a global optimum!)
Iterated Local Search (ILS) tries to search through this space of local optima in a more intelligent
fashion: it tries to stochastically hill-climb in the space of local optima. That is, ILS finds a local optimum,
then looks for a “nearby” local optimum and possibly adopts that one instead, then finds a new
“nearby” local optimum, and so on. The heuristic here is that you can often find better local optima
near to the one you’re presently in, and walking from local optimum to local optimum in this way
often outperforms just trying new locations entirely at random.
ILS pulls this off with two tricks. First, ILS doesn’t pick new restart locations entirely at random.
Rather, it maintains a “home base” local optimum of sorts, and selects new restart locations that
are somewhat, though not excessively, in the vicinity of the “home base” local optimum. We want
to restart far enough away from our current home base to wind up in a new local optimum, but
not so far as to be picking new restart locations essentially at random. We want to be doing a walk
rather than a random search.
Second, when ILS discovers a new local optimum, it decides whether to retain the current
“home base” local optimum, or to adopt the new local optimum as the “home base”. If we always
pick the new local optimum , we’re doing a random walk (a sort of meta-exploration). If we only
pick the new local optimum if it’s better than our current one, we’re doing hill-climbing (a sort of
meta-exploitation). ILS often picks something in-between the two, as discussed later.
If you abstract these two tricks, ILS is very simple. The only complexity lies in determining
when a local optimum has been discovered. Since this is often difficult, I will instead employ the
same approach here as was used in random restarts: to set a timer. Hill-climb for a while, and then
when timer goes off, it’s time to restart. This obviously doesn’t guarantee that we’ve found the
local optimum while hill-climbing, but if the timer is long enough, we’re likely to be in the vicinity.
The algorithm is very straightforward: do hill-climbing for a while; then (when time is up)
determine whether to adopt the newly discovered local optimum or to retain the current “home
base” one (the NewHomeBase12 function); then from our new home base, make a very big Tweak
(the Perturb function), which is ideally just large enough to likely jump to a new hill. The algorithm
looks like this:
11 A good current summary of the technique can be found in Helena Lourenço, Olivier Martin, and Thomas Stützle,
2003, Iterated local search, in Fred Glover and Gary Kochenberger, editors, Handbook of Metaheuristics, pages 320–353,
Springer. They trace the technique back as far as John Baxter, 1981, Local optima avoidance in depot location, Journal of
the Operational Research Society, 32, 815–819.
12 I made up that name.
26
Algorithm 16 Iterated Local Search (ILS) with Random Restarts
2: S ← some initial random candidate solution

3: H←S " The current “home base” local optimum
4: Best ← S
5: repeat
7: repeat
10: S←R
11: until S is the ideal solution, or time is up, or we have run out of total time
13: Best ← S
14: H ← NewHomeBase(H, S)
15: S ← Perturb(H)
16: until Best is the ideal solution or we have run out of total time
17: return Best
Much of the thinking behind the choices of Perturb and NewHomeBase functions is a black art,
determined largely by the nature of the particular problem being tackled. Here are some hints.
The goal of the Perturb function is to make a very large Tweak, big enough to likely escape the
current local optimum, but not so large as to be essentially a randomization. Remember that we’d
like to fall onto a nearby hill. The meaning of “big enough” varies wildly from problem to problem.
The goal of the NewHomeBase function is to intelligently pick new starting locations. Just as
global optimization algorithms in general lie between the extremes of exploration (random search
and random walks) and exploitation (hill-climbing), the NewHomeBase should lie somewhere
between these extremes when considering among local optima.13 At one extreme, the algorithm
could always adopt the new local optimum, that is,
NewHomeBase( H, S) = S
This results in essentially a random walk from local optimum to local optimum. At the other
extreme, the algorithm could only use the new local optimum if it’s of equal or higher quality than
the old one, that is,
(
S if Quality(S) ≥ Quality( H )
NewHomeBase( H, S) =
H otherwise
This results, more or less, in a kind of hill-climbing among the local optima. Most ILS heuristics try
to strike a middle-ground between the two. For example, ILS might hill-climb unless it hasn’t seen
a new and better solution in a while, at which point it starts doing random walks for a bit. There
are other options of course: we could apply a Simulated Annealing approach to NewHomeBase, or a
Tabu Search procedure of sorts.
13 Thus this function truly is a meta-heuristic. Finally a valid use of the term!
27
Mixing and Matching The algorithms described in this section are not set in stone. There are lots
of ways to mix and match them, or other approaches entirely. For example, it’s not unreasonable to
use Hill-Climbing with Random Restarts mixed with a (1 + 1)-style Tweak operation. You could
also construct Steepest Ascent versions of Random Restarts. Tabu Search could be done in (1, λ)
style. Or construct a Tweak procedure which slowly decreases Gaussian convolution’s σ2 according
to a Simulated Annealing-style temperature. And so on. Be imaginative.
28
3 Population Methods
Population-based methods differ from the previous methods in that they keep around a sample of
candidate solutions rather than a single candidate solution. Each of the solutions is involved in
tweaking and quality assessment, but what prevents this from being just a parallel hill-climber is
that candidate solutions affect how other candidates will hill-climb in the quality function. This
could happen either by good solutions causing poor solutions to be rejected and new ones created,
or by causing them to be Tweaked in the direction of the better solutions.
It may not be surprising that most population-based methods steal concepts from biology.
One particularly popular set of techniques, collectively known as Evolutionary Computation
(EC), borrows liberally from population biology, genetics, and evolution. An algorithm chosen
from this collection is known as an Evolutionary Algorithm (EA). Most EAs may be divided
into generational algorithms, which update the entire sample once per iteration, and steady-state
algorithms, which update the sample a few candidate solutions at a time. Common EAs include
the Genetic Algorithm (GA) and Evolution Strategies (ES); and there are both generational and
steady-state versions of each. There are quite a few more alphabet soup subalgorithms.
Because they are inspired by biology, EC methods tend to use (and abuse) terms from genetics
and evolution. Because the terms are so prevalent, we’ll use them in this and most further sections.
Definition 1 Common Terms Used in Evolutionary Computation

individual a candidate solution
child and parent a child is the Tweaked copy of a candidate solution (its parent)
population set of candidate solutions
fitness quality
fitness landscape quality function
fitness assessment or evaluation computing the fitness of an individual
selection picking individuals based on their fitness
mutation plain Tweaking. This is often thought as “asexual” breeding.
recombination or crossover A special Tweak which takes two parents, swaps sections of
them, and (usually) produces two children. This is often
thought as “sexual” breeding.
breeding producing one or more children from a population of parents
through an iterated process of selection and Tweaking (typically
mutation or recombination)
genotype or genome an individual’s data structure, as used during breeding
chromosome a genotype in the form of a fixed-length vector
gene a particular slot position in a chromosome
allele a particular setting of a gene
phenotype how the individual operates during fitness assessment
generation one cycle of fitness assessment, breeding, and population re-
assembly; or the population produced each such cycle
Evolutionary Computation techniques are generally resampling techniques: new samples

(populations) are generated or revised based on the results from older ones. In contrast, Parti-
cle Swarm Optimization, in Section 3.6, is an example of a directed mutation method, where
candidate solutions in the population are modified, but no resampling occurs per se.
29
The basic generational evolutionary computation algorithm first constructs an initial population,
then iterates through three procedures. First, it assesses the fitness of all the individuals in the
population. Second, it uses this fitness information to breed a new population of children. Third, it
joins the parents and children in some fashion to form a new next-generation population, and the
cycle continues.
Algorithm 17 An Abstract Generational Evolutionary Algorithm (EA)

1: P ← Build Initial Population
2: Best ← ! " ! means “nobody yet”
3: repeat
4: AssessFitness(P)
5: for each individual Pi ∈ P do
6: if Best = ! or Fitness(Pi ) > Fitness(Best) then " Remember, Fitness is just Quality
7: Best ← Pi
8: P ← Join(P, Breed(P))
10: return Best
Notice that, unlike the Single-State methods, we now have a separate AssessFitness function.
This is because typically we need all the fitness values of our individuals before we can Breed them.
So we have a certain location in the algorithm where their fitnesses are computed.
Evolutionary algorithms differ from one another largely in how they perform the Breed and Join
operations. The Breed operation usually has two parts: Selecting parents from the old population,
then Tweaking them (usually Mutating or Recombining them in some way) to make children.
The Join operation usually either completely replaces the parents with the children, or includes fit
parents along with their children to form the next generation.14
Population Initialization All the algorithms described here basically use the same initialization
procedures, so it’s worthwhile giving some tips. Initialization is typically just creating some n
individuals at random. However, if you know something about the likely initial “good” regions of
the space, you could bias the random generation to tend to generate individuals in those regions. In
fact, you could seed the initial population partly with individuals of your own design. Be careful
about such techniques: often though you think you know where the good areas are, there’s a good
chance you don’t. Don’t put all your eggs in one basket: include a significant degree of uniform
randomness in your initialization. More on this later on when we talk about representations (in
Section 4.1.1).
It’s also worthwhile to enforce diversity by guaranteeing that every individual in the initial
population is unique. Each time you make a new individual, don’t scan through the whole
population to see if that individual’s already been created: that’s O(n2 ) and foolish. Instead, create
a hash table which stores individuals as keys and anything arbitrary as values. Each time you make
an individual, check to see if it’s already in the hash table as a key. If it is, throw it away and make
another one. Else, add the individual to the population, and hash it in the hash table. That’s O(n).
14 Though it’s usually simpler than this, the Join operation can be thought of as kind of selection procedure, choosing
from among the children and the parents to form the next generation. This general view of the Join operation is often
called survival selection, while the selection portion of the Breed operation is often called parent selection.
30
3.1 Evolution Strategies
The family of algorithms known as Evolution Strategies (ES) were developed by Ingo Rechenberg
and Hans-Paul Schwefel at the Technical University of Berlin in the mid 1960s.15 ES employ a
simple procedure for selecting individuals called Truncation Selection, and (usually) only uses
mutation as the Tweak operator.
Among the simplest ES algorithms is the (µ, λ) algorithm. We begin with a population of
(typically) λ number of individuals, generated randomly. We then iterate as follows. First we assess
the fitness of all the individuals. Then we delete from the population all but the µ fittest ones (this
is all there’s to Truncation Selection). Each of the µ fittest individuals gets to produce λ/µ children
through an ordinary Mutation. All told we’ve created λ new children. Our Join operation is simple:
the children just replace the parents, who are discarded. The iteration continues anew.
In short, µ is the number of parents which survive, and λ is the number of kids that the µ
parents make in total. Notice that λ should be a multiple of µ. ES practitioners usually refer to
their algorithm by the choice of µ and λ. For example, if µ = 5 and λ = 20, then we have a (5, 20)
Evolution Strategy. Here’s the algorithm pseudocode:
Algorithm 18 The (µ, λ) Evolution Strategy

1: µ ← number of parents selected
2: λ ← number of children generated by the parents
3: P ← {}
4: for λ times do " Build Initial Population
5: P ← P ∪ {new random individual}
6: Best ← !
7: repeat
9: AssessFitness(Pi )
10: if Best = ! or Fitness(Pi ) > Fitness(Best) then
11: Best ← Pi
12: Q ← the µ individuals in P whose Fitness( ) are greatest " Truncation Selection
13: P ← {} " Join is done by just replacing P with the children
14: for each individual Q j ∈ Q do
15: for λ/µ times do
16: P ← P ∪ {Mutate(Copy(Q j ))}
18: return Best
Note the use of the function Mutate instead of Tweak. Recall that population-based methods
have a variety of ways to perform the Tweak operation. The big two are mutation, which is just
like the Tweaks we’ve seen before: convert a single individual into a new individual through a
(usually small) random change; and recombination or crossover, in which multiple (typically two)
individuals are mixed and matched to form children. We’ll be using these terms in the algorithms
from now on out to indicate the Tweak performed.
15 Ingo
Rechenberg, 1973, Evolutionsstrategie: Optimierung technischer Systeme nach Prinzipien der biologischen Evolution,
Fromman-Holzbook, Stuttgart, Germany. In German!
31
The (µ, λ) algorithm has three knobs with which we may adjust exploration versus exploitation.
Figure 8 shows the effect of variations with these operations.
• The size of λ. This essentially controls the sample size for each population, and is basically
the same thing as the n variable in Steepest-Ascent Hill Climbing With Replacement. At the
extreme, as λ approaches ∞, the algorithm approaches exploration (random search).
• The size of µ. This controls how selective the algorithm is; low values of µ with respect to λ
push the algorithm more towards exploitative search as only the best individuals survive.
• The degree to which Mutation is performed. If Mutate has a lot of noise, then new children
fall far from the tree and are fairly random regardless of the selectivity of µ.
The second Evolution Strategy algorithm is called (µ + λ). It differs from (µ, λ) in only one
respect: the Join operation. Recall that in (µ, λ) the parents are simply replaced with the children
in the next generation. But in (µ + λ), the next generation consists of the µ parents plus the λ
new children. That is, the parents compete with the kids next time around. Thus the next and all
successive generations are µ + λ in size. The algorithm looks like this:
Algorithm 19 The (µ + λ) Evolution Strategy

1: µ ← number of parents selected
2: λ ← number of children generated by the parents
3: P ← {}
4: for λ times do
6: Best ← !
7: repeat
11: Best ← Pi
12: Q ← the µ individuals in P whose Fitness( ) are greatest
13: P←Q " The Join operation is the only difference with (µ, λ)
15: for λ/µ times do
16: P ← P ∪ {Mutate(Copy(Q j ))}
18: return Best
Generally speaking, (µ + λ) may be more exploitative than (µ, λ) because high-fit parents
persist to compete with the children. This has risks: a sufficiently fit parent may defeat other
population members over and over again, eventually causing the entire population to prematurely
converge to immediate descendants of that parent, at which point the whole population has been
trapped in the local optimum surrounding the parent.
32
Breeding Operations
Generation 4
Generation 3
Generation 2
(1, 2) Evolution Strategy
Generation 1
Individuals Selected To Breed
Individuals Not Selected
Figure 8 Three (µ, λ) Evolution Strategy variations. Each generation, µ individuals are selected to breed, and each gets
to create λ/µ children, resulting in λ children in total.
If you think about it, (µ + λ) resembles Steepest Ascent Hill-Climbing in that both of them
allow the parent to compete against the children for supremacy in the next iteration. Whereas
(µ, λ) resembles Steepest Ascent Hill-Climbing with Replacement in that the parents are replaced
with the best children. This is more than a coincidence: the hill-climbers are essentially degenerate
cases of ES algorithms. Recall that with the right Tweak operator, plain Hill-Climbing becomes the
(1 + 1) algorithm, Steepest Ascent Hill-Climbing with Replacement becomes (1, λ), and Steepest
Ascent Hill-Climbing becomes (1 + λ). Armed with the explanation of the algorithms above, it
should be a bit clearer why this is.
3.1.1 Mutation and Evolutionary Programming

Evolution Strategies historically employ a representation in the form of a fixed-length vector of
real-valued numbers. Typically such vectors are initialized using something along the lines of
Algorithm 7. Mutation is typically performed using Gassian Convolution (Algorithm 11).
Gaussian Convolution is controlled largely by the distribution variance σ2 . The value of σ2
is known as the mutation rate of an ES, and determines the noise in the Mutate operation. How
do you pick a value for σ2 ? You might pre-select its value; or perhaps you might slowly decrease
the value; or you could try to adaptively change σ2 based on the current statistics of the system.
If the system seems to be too exploitative, you could increase σ2 to force some more exploration
(or likewise decrease it to produce more exploitation). This notion of changing σ2 is known as an
adaptive mutation rate. In general, such adaptive breeding operators adjust themselves over time,
in response to statistics gleaned from the optimization run.16
16 Evolution Strategies have also long been associated with self-adaptive operators which are stochastically optimized
along with individuals. For example, individuals might contain their own mutation procedures which can themselves
be mutated along with the individual.
33
One old rule for changing σ2 adaptively is known as the One-Fifth Rule, by Ingo Rechenberg,17
and it goes like this:
• If more than 15 children are fitter than their parents, then we’re exploiting local optima too
much, and we should increase σ2 .
• If less than 15 children are fitter than their parents, then we’re exploring too much, and we
should decrease σ2 .
1
• If exactly 5 children are fitter than their parents, don’t change anything.
This rule was derived from the results of experiments with the (1 + 1) ES on certain simple test
problems. It may not be optimal for more complex situations: but it’s a good starting point.
You don’t have to do ES just with vectors. In fact, a little earlier than ES, an almost identical
approach was developed by Larry Fogel at the National Science Foundation (Washington DC) and
later developed in San Diego.18 The technique, called Evolutionary Programming (EP), differs
from ES in two respects. First, it historically only used a (µ + λ) strategy with µ = λ. That is, half
the population was eliminated, and that half was then filled in with children. Second, EP was
applied to most any representation. From the very start Fogel was interested in evolving graph
structures (specifically finite state automata, hence the “programming”). Thus the Mutate operation
took the form of adding or deleting an edge, adding or deleting a node, relabeling an edge or a
node, etc.
Such operations are reasonable as long as they have two features. First, to guarantee that
the algorithm remains global, we must guarantee that, with some small probability, a parent can
produce any child. Second, we ought to retain the feature that usually we make small changes likely
to not deviate significantly in fitness; and only occasionally make large changes to the individual.
The degree to which we tend to make small changes could be adjustable, like σ2 was. We’ll get to
such representational issues for candidate solutions in detail in Section 4.
3.2 The Genetic Algorithm

The Genetic Algorithm (GA), often referred to as genetic algorithms, was invented by John Holland
at the University of Michigan in the 1970s.19 It is similar to a (µ, λ) Evolution Strategy in many
respects: it iterates through fitness assessment, selection and breeding, and population reassembly.
The primary difference is in how selection and breeding take place: whereas Evolution Strategies
select the parents and then creates the children, the Genetic Algorithm little-by-little selects a few
parents and generates children until enough children have been created.
To breed, we begin with an empty population of children. We then select two parents from the
original population, copy them, cross them over with one another, and mutate the results. This
forms two children, which we then add to the child population. We repeat this process until the
child population is entirely filled. Here’s the algorithm in pseudocode.
17 Also in his evolution strategies text (see Footnote 15, p. 31).
18 Larry Fogel’s dissertation was undoubtedly the first such thesis, if not the first major work, in the field of evolutionary
computation. Lawrence Fogel, 1964, On the Organization of Intellect, Ph.D. thesis, University of California, Los Angeles.
19 Holland’s book is one of the more famous in the field: John Holland, 1975, Adaptation in Natural and Artificial Systems,
University of Michigan Press.
34
Algorithm 20 The Genetic Algorithm (GA)
1: popsize ← desired population size " This is basically λ. Make it even.
2: P ← {}
3: for popsize times do
5: Best ← !
6: repeat
10: Best ← Pi
11: Q ← {} " Here’s where we begin to deviate from (µ, λ)
12: for popsize/2 times do
13: Parent Pa ← SelectWithReplacement(P)
14: Parent Pb ← SelectWithReplacement(P)
15: Children Ca , Cb ← Crossover(Copy(Pa ), Copy(Pb ))
16: Q ← Q ∪ {Mutate(Ca ), Mutate(Cb )}
17: P←Q " End of deviation
19: return Best
Though it can be applied to any kind of vector (and indeed many representations) the GA
classically operated over fixed-length vectors of boolean values, just like ES often were applied to
ones of floating-point values. For a moment, let’s be pedantic about generation of new individuals. If
the individual is a vector of floating-point values, creating a new random vector could be done just
like in ES (that is, via Algorithm 7). If our representation is a boolean vector, we could do this:
Algorithm 21 Generate a Random Bit-Vector

v ← a new vector $v1 , v2 , ...vl %
1: !
3: if 0.5 > a random number chosen uniformly between 0.0 and 1.0 inclusive then
4: vi ← true
5: else
6: vi ← false
7: return ! v
3.2.1 Crossover and Mutation

Note how similar the Genetic Algorithm is to (µ, λ), except during the breeding phase. To perform
breeding, we need two new functions we’ve not seen before: SelectWithReplacement and Crossover;
plus of course Mutate. We’ll start with Mutate. Mutating a real-valued vector could be done with
Gaussian Convolution (Algorithm 11). How might you Mutate a boolean vector? One simple way is
35
bit-flip mutation: march down the vector, and flip a coin of a certain probability (often 1/l, where
l is the length of the vector). Each time the coin comes up heads, flip the bit:
Algorithm 22 Bit-Flip Mutation

1: p ← probability of flipping a bit " Often p is set to 1/l
v ← boolean vector $v1 , v2 , ...vl % to be mutated
2: !
4: if p ≥ random number chosen uniformly from 0.0 to 1.0 inclusive then
5: vi ← ¬(vi )
6: return !v
Crossover is the Genetic Algorithm’s distinguishing

feature.20 It involves mixing and matching parts of two
1 1 0 0 1 0 0 1
parents to form children. How you do that mixing and
matching depends on the representation of the individuals. Swap Swap Swap
There are three classic ways of doing crossover in vectors:

One-Point, Two-Point, and Uniform Crossover. 0 0 1 0 1 1 0 0
Let’s say the vector is of length l. One-point crossover
picks a number c between 1 and l, inclusive, and swaps c
all the indexes greater than c, as shown in Figure 9. The
Figure 9 One-Point Crossover.
algorithm:
Algorithm 23 One-Point Crossover

v ← first vector $v1 , v2 , ...vl % to be crossed over
1: !
! ← second vector $w1 , w2 , ...wl % to be crossed over
2: w
3: c ← random integer chosen uniformly from 1 to l inclusive

4: for i from c to l do
5: Swap the values of vi and wi
6: return ! v and w
!
Note that if c = 1 or c = l, no crossover really happens:

if c = 1 then everything crosses over and if c = l nothing
1 1 0 0 1 0 0 1
crosses over. In either case, the individuals don’t change.
The problem with one-point crossover lies in the pos- Swap Swap Swap
sible linkage (also called epistasis) among the elements in

the vector. Notice that the probability is high that v1 and 0 0 1 0 1 1 0 0
vl will be broken up due to crossover, as almost any choice
of c will do it. Similarly, the probability that v1 and v2 will d c
be broken up is quite small, as c must be equal to 1. If the
Figure 10 Two-Point Crossover.
organization of your vector was such that elements v1 and
vl had to work well in tandem in order to get a high fitness,
you’d be constantly breaking up good pairs that the system discovered. Two-point crossover is one
20 Though it’s long since been used in various ways with Evolution Strategies as well.
36
way to clean up the linkage problem: just pick two numbers c and d, and swap the indexes between
them. Figure 10 gives the general idea, and the pseudocode is below:
Algorithm 24 Two-Point Crossover

1: !
2: w

4: d ← random integer chosen uniformly from 1 to l inclusive
5: if c > d then
6: Swap c and d
7: for i from c to d − 1 do
9: return !v and w
!
It’s not immediately obvious how this would help things. But think of the vectors not as vectors
but as rings (that is, vl is right next to v1 ). Two-point crossover breaks the rings at two spots and
trades pieces. Since vl is right next to v1 , the only way they’d break up is if c or d sliced right
between them. The same situation as v1 and v2 .21
Even so, there’s still a further linkage problem. v1 and
vl are now being treated fairly, but how about v1 and vl/2 ?
1 1 0 0 1 0 0 1
Long distances like that are still more likely to be broken
up than short distances like v1 and v2 (or indeed v1 and Swap Swap Swap Swap
vl ). We can treat all genes fairly with respect to linkage

by crossing over each point independently of one another, 0 0 1 0 1 1 0 0
using Uniform crossover. Here we simply march down the
vectors, and swap individual indexes if a coin toss comes Figure 11 Uniform Crossover.
up heads with probability p. 22
Algorithm 25 Uniform Crossover

1: p ← probability of swapping an index " Often p is set to 1/l. At any rate, p ≤ 0.5
2: !
3: w
7: return ! v and w
!
21 We can generalize two-point crossover into a Multi-Point Crossover: pick n random points and sort them smallest
first: c1 , c2 , ..., cn . Now swap indexes in the region between c1 and c2 , and between c3 and c4 , and likewise c5 and c6 , etc.
22 The original uniform crossover assumed p = 1/2, and was first proposed in David Ackley, 1987, A Connectionist
Machine for Genetic Hillclimbing, Kluwer Academic Publishers. The more general form, for arbitrary p, is sometimes
called parameterized uniform crossover.
37
Crossover is not a global mutation. If you cross over two vectors you can’t get every conceivable
vector out of it. Imagine your vectors were points in space. Now imagine the hypercube formed
with those points at its extreme corners. For example, if your vectors were 3-dimensional, they’d
form the corners of a cube in space, as shown in Figure 12. All the crossovers so far are very
constrained: they will result in new vectors which lie at some
other corner of the hypercube.
By extension, imagine an entire population P as points
in space (such as the three-dimensional space in Figure 12).
Crossover done on P can only produce children inside the
bounding box surrounding P in space. Thus P) s bounding box
can never increase: you’re doomed to only search inside it.
As we repeatedly perform crossover and selection on a pop-
ulation, it may reach the situation where certain alleles (values
for certain positions in the vector) have been eliminated, and
the bounding box will collapse in that dimension. Eventually
the population will converge, and often (unfortunately) pre-
maturely converge, to copies of the same individual. At this
stage there’s no escape: when an individual crosses over with Figure 12 A cube in space formed by two
itself, nothing new is generated.23 Thus to make the Genetic three-dimensional vectors (black circles).
The dashed line connects the two vectors.
Algorithm global, you also need to have a Mutate operation.
What’s the point of crossover then? Crossover was origi-
nally based on the premise that highly fit individuals often share certain traits, called building
blocks, in common. For fixed-length vector individuals a building block was often defined as a
collection of genes set to certain alleles. For example, in the boolean individual 10110101, perhaps
***101*1 might be a building block (where the * positions aren’t part of the building block). In many
problems for which crossover was helpful, the fitness of a given individual is often at least partly
correlated to the degree to which it contains various of these building blocks, and so crossover
works by spreading building blocks quickly throughout the population. Building blocks were the
focus of much early genetic algorithm analysis, formalized in an area known as schema theory.
That’s the idea anyway. But, hand-in-hand with this building-block hypothesis, Crossover
methods also assume that there is some degree of linkage24 between genes on the chromosome: that
is, settings for certain genes in groups are strongly correlated to fitness improvement. For example,
genes A and B might contribute to fitness only when they’re both set to 1: if either is set to 0, then
the fact that the other is set to 1 doesn’t do anything. One- and Two-point Crossover also make
the even more tenuous assumption that your vector is structured such that highly linked genes
are located near to one another on the vector: because such crossovers are unlikely to break apart
closely-located gene groups. Unless you have carefully organized your vector, this assumption
is probably a bug, not a feature. Uniform Crossover also makes some linkage assumptions but
does not have this linkage-location bias. Is the general linkage assumption true for your problem?
Or are your genes essentially independent of one another? For most problems of interest, it’s the
former: but it’s dicey. Be careful.
23 Crossovers which don’t make anything new when an individual crosses over with itself are called homologous.
24 One special kind of linkage effect has its own term stolen straight from biology: epistasis. Here, genes A and B
are linked because gene B has an effect on the expression of gene A (on the other hand, A may not affect B). The term
“epistasis” can also be used more generally as a synonym for linkage.
38
In theory, you could perform uniform crossover with several vectors at once to produce children
which are the combination of all of them.25 To avoid sheer randomization, probably you’d want only
a bit of mixing to occur, so the probability of swapping any given index shouldn’t be spectacularly
high. Something like this is very rare in practice though. To do it, we first need to define how to
uniformly randomly shuffle a vector. Surprisingly, it’s not as obvious as you’d think.
Algorithm 26 Randomly Shuffle a Vector

1: !p ← elements to shuffle $ p1 , ..., pl %
2: for i from l down to 2 do " Note we don’t do 1

3: j ← integer chosen at random from 1 to i inclusive
4: Swap pi and p j
Armed with a random shuffler (we’ll use it in future algorithms too), we can now cross over k
vectors at a time, trading pieces with one another, and producing k children as a result.
Algorithm 27 Uniform Crossover among K Vectors

1: p ← probability of swapping an index " Ought to be very small
2: W ← {W1 , ..., Wk } vectors to cross over, each of length l
3: !v ← vector $v1 , ..., vk %

6: for j from 1 to k do " Load !v with the ith elements from each vector in W
7: w
! ← Wj
8: v j ← wi
9: Randomly Shuffle !v
10: for j from 1 to k do " Put back the elements, all mixed up
11: w
! ← Wj
12: wi ← v j
13: Wj ← w!
14: return W
3.2.2 More Recombination

So far we’ve been doing crossovers that are just swaps: but if the vectors are of floating-point
values, our recombination could be something fuzzier, like averaging the two values rather than
swapping them. Imagine if our two vectors were points in space. We draw a line between the two
points and choose two new points between them. We could extend this line somewhat beyond the
points as well, as shown in the dashed line in Figure 12, and pick along the line. This algorithm,
known as Line Recombination, here presented in the form given by Heinz Mühlenbein and Dirk
Schlierkamp-Voosen, depends on a variable p which determines how far out along the line we’ll
allow children to be. If p = 0 then the children will be located along the line within the hypercube
(that is, between the two points). If p > 0 then the children may be located anywhere on the line,
even somewhat outside of the hypercube.
25 There’s nothing new under the sun: this was one of the early ES approaches tried by Hans-Paul Schwefel.
39
Algorithm 28 Line Recombination
1: p ← positive value which determines how far long the line a child can be located " Try 0.25
2: !
3: w
4: α ← random value from − p to 1 + p inclusive

5: β ← random value from − p to 1 + p inclusive
7: t ← αvi + (1 − α)wi
8: s ← βwi + (1 − β)vi
9: if t and s are within bounds then
10: vi ← t
11: wi ← s
12: return !v and w!
We could extend this further by picking random α and β values for each position in the vector.
This would result in children that are located within the hypercube or (if p > 0) slightly outside of
it. Mühlenbein and Schlierkamp-Voosen call this Intermediate Recombination.26
Algorithm 29 Intermediate Recombination

1: p ← positive value which determines how far long the line a child can be located " Try 0.25
2: !
3: w
5: repeat
6: α ← random value from − p to 1 + p inclusive " We just moved these two lines!
8: t ← αvi + (1 − α)wi
9: s ← βwi + (1 − β)vi
10: until t and s are within bounds
11: vi ← t
12: wi ← s
Since we’re using different values of α and β for each element, instead of rejecting recombination
if the elements go out of bounds, we can now just repeatedly pick a new α and β.
Why bother with values of p > 0? Imagine that you have no Mutate operation, and are just
doing Intermediate or Line Recombination. Each time you select parents and generate a child,
26 Okay, they called them Extended Line and Extended Intermediate Recombination, in Heinz Mühlenbein and Dirk
Schlierkamp-Voosen, 1993, Predictive models for the breeder genetic algorithm: I. continuous parameter optimization,
Evolutionary Computation, 1(1). These methods have long been in evolutionary computation, but the terms are hardly
standardized: notably Hans-Paul Schwefel’s original Evolutionary Strategies work used (among others) line recombina-
tion with p = −0.5, but he called it intermediate recombination, as do others. Schwefel also tried a different variation: for
each gene of the child, two parents were chosen at random, and their gene values at that gene were averaged.
40
that child is located somewhere within the cube formed by the parents (recall Figure 12). Thus it’s
impossible to generate a child outside the bounding box of the population. If you want to explore in
those unknown regions, you need a way to generate children further out than your parents are.
Other Representations So far we’ve focused on vectors. In Section 4 we’ll get to other repre-
sentations. For now, remember that if you can come up with a reasonable notion of Mutate, any
representation is plausible. How might we do graph structures? Sets? Arbitrary-length lists? Trees?
3.2.3 Selection
In Evolution Strategies, we just lopped off all but the µ best individuals, a procedure known as
Truncation Selection. Because the Genetic Algorithm performs iterative selection, crossover, and
mutation while breeding, we have more options. The GA’s SelectWithReplacement procedure will
select an individual in the population such that, in the future, there’s no reason it couldn’t by chance
happen to select the same individual again. Whereas in an ES an individual will be the parent of a
predefined number of children, in the GA an individual can have any number of children.
The original technique for GA selec-
tion was called Fitness-Proportionate Se- Total Fitness Range 0 s
lection, sometimes known as Roulette Se-
Individuals
lection. In this algorithm, we select indi- Sized by Fitness
1 2 3 4 5 6 7 8
viduals in proportion to their fitness: if an

individual has a higher fitness, it’s selected Figure 13 Array of individual ranges in Fitness Proportionate
Selection.
more often.27 To do this we “size” the indi-
viduals according to their fitness as shown
in Figure 13.28 Let s = ∑i f i be the sum fitness of all the individuals. A random number from 0 to s
falls within the range of some individual, which is then selected.
Algorithm 30 Fitness-Proportionate Selection

1: perform once per generation
2: global !p ← population copied into a vector of individuals $ p1 , p2 , ..., pl %
3: global !f ← $ f 1 , f 2 , ..., f l % fitnesses of individuals in !p in the same order as !p " Must all be ≥ 0
4: if !f is all 0.0s then " Deal with all 0 fitnesses gracefully
5: Convert !f to all 1.0s
6: for i from 2 to l do " Convert !f to a CDF. This will also cause f l = s, the sum of fitnesses.
7: f i ← f i + f i −1
8: perform each time
9: n ← random number from 0 to f l inclusive
10: for i from 2 to l do " This could be done more efficiently with binary search
11: if f i−1 < n ≤ f i then
12: return pi
13: return p1
27 We presume here that fitnesses are ≥ 0. As usual, higher is better.

28 Also due to John Holland. See Footnote 19, p. 34.
41
Notice that Fitness-Proportionate Selection has a preprocessing step: converting all the fitnesses
(or really copies of them) into a cumulative distribution. This only needs to be done once per
generation. Additionally, though the code searches linearly through the fitness array to find the
one we want, it’d be smarter to do that in O(lg n) time by doing a binary search instead.
One variant on Fitness-Proportionate Se-
lection is called Stochastic Universal Sam- Total Fitness Range 0 s
pling (or SUS), by James Baker. In SUS,
Individuals
we select in a fitness-proportionate way but Sized by Fitness
1 2 3 4 5 6 7 8
biased so that fit individuals always get Start Range! 0

(here n = 8)!
s/n!
picked at least once. This is known as a low n Chosen Individuals

variance resampling algorithm and I include Begins within the Start Range 1 1 3 4 5 5 7 8
it here because it is now popular in other

Figure 14 Array of individual ranges, start range, and chosen
venues than just evolutionary computation points in Stochastic Universal Sampling.
(most famously, Particle Filters for Markov
Decision Processes (MDPs)).29
SUS selects n individuals at a time (typically n is the size of the next generation, so in our case
n = l). To begin, we build our fitness array as before. Then we select a random position from 0 to
s/n. We then select the individual which straddles that position. We then increment the position
by s/n and repeat (up to n times total). Each increment, we select the individual in whose fitness
region we landed. This is shown in Figure 14. The algorithm is:
Algorithm 31 Stochastic Universal Sampling

1: perform once per n individuals produced " Usually n = l, that is, once per generation
2: global !p ← copy of vector of individuals (our population) $ p1 , p2 , ..., pl %, shuffled randomly
" To shuffle a vector randomly, see Algorithm 26
3: !
global f ← $ f 1 , f 2 , ..., f l % fitnesses of individuals in !p in the same order as !p " Must all be ≥ 0
4: global index ← 0
5: if !f is all 0.0s then
6: Convert !f to all 1.0s
7: for i from 2 to l do " Convert !f to a CDF. This will also cause f l = s, the sum of fitnesses.
8: f i ← f i + f i −1
9: global value ← random number from 0 to f l /n inclusive
11: while f index < value do
12: index ← index + 1
13: value ← value + f l /n
14: return pindex
There are basically two advantages to SUS. First, it’s O(n) to select n individuals, rather than
O(n lg n) for Fitness-Proportionate Selection. That used to be a big deal but it isn’t any more,
since the lion’s share of time in most optimization algorithms is spent in assessing the fitness
29 And they never seem to cite him.Here it is: James Edward Baker, 1987, Reducing bias and inefficiency in the selection
algorithm, in John Grefenstette, editor, Genetic Algorithms and Their Applications: Proceedings of the Second International
Conference on Genetic Algorithms (ICGA), pages 14–21, Lawrence Erlbaum Associates, Hillsdale.
42
of individuals, not in the selection or breeding processes. Second and more interesting, SUS
guarantees that if an individual is fairly fit (over s/n in size), it’ll get chosen for sure, sometimes
multiple times. In Fitness-Proportionate Selection even the fittest individual may never be selected.
There is a big problem with the methods described so far: they presume that the actual fitness
value of an individual really means something important. But often we choose a fitness function
such that higher ones are “better” than smaller ones, and don’t mean to imply anything else. Even if
the fitness function was carefully chosen, consider the following situation, where a fitness function
goes from 0 to 10. Near the end of a run, all the individuals have values like 9.97, 9.98, 9.99, etc. We
want to finesse the peak of the fitness function, and so we want to pick the 9.99-fitness individual.
But to Fitness-Proportionate Selection (and to SUS), all these individuals will be selected with
nearly identical probability. The system has converged to just doing random selection.
To fix this we could scale the fitness function to be more sensitive to the values at the top end
of the function. But to really remedy the situation we need to adopt a non-parametric selection
algorithm which throws away the notion that fitness values mean anything other than bigger is
better, and just considers their rank ordering. Truncation Selection does this, but the most popular
technique by far is Tournament Selection,30 an astonishingly simple algorithm:
Algorithm 32 Tournament Selection

1: P ← population
2: t ← tournament size, t ≥ 1
3: Best ← individual picked at random from P with replacement

4: for i from 2 to t do
5: Next ← individual picked at random from P with replacement
6: if Fitness(Next) > Fitness(Best) then
7: Best ← Next
8: return Best
We return the fittest individual of some t individuals picked at random, with replacement, from
the population. That’s it! Tournament Selection has become the primary selection technique used
for the Genetic Algorithm and many related methods, for several reasons. First, it’s not sensitive
to the particulars of the fitness function. Second, it’s dead simple, requires no preprocessing, and
works well with parallel algorithms. Third, it’s tunable: by setting the tournament size t, you
can change how selective the technique is. At the extremes, if t = 1, this is just random search.
If t is very large (much larger than the population size itself), then the probability that the fittest
individual in the population will appear in the tournament approaches 1.0, and so Tournament
Selection just picks the fittest individual each time (put another way, it approaches Truncation
Selection with µ = 1).
In the Genetic Algorithm, the most popular setting is t = 2. For certain representations (such as
those in Genetic Programming, discussed later in Section 4.3), it’s common to be more selective
(t = 7). To be less selective than t = 2, but not be totally random, we’d need some kind of trick. One
way I do it is to also allow real-numbered values of t from 1.0 to 2.0. In this range, with probability
t − 1.0, we do a tournament selection of size t = 2, else we select an individual at random (t = 1).31
30 Tournament Selection may be a folk algorithm: but the earliest usage I’m aware of is Anne Brindle, 1981, Genetic
Algorithms for Function Optimization, Ph.D. thesis, University of Alberta. She used binary tournament selection (t = 2).
31 You could generalize this to any real-valued t ≥ 1.0: with probability t − 0 t 1 select with size 2 t 3, else with size 0 t 1.
43
3.3 Exploitative Variations
It seems the trend in new algorithms is to be more exploitative. Some variations such as Elitism, the
Steady-State Genetic Algorithm (and Generation Gap methods), and the Genetic Algorithm with
a Tree-Style Genetic Programming Pipeline, are exploitative because highly-fit parents can linger
in the population and compete with their children, like (µ + λ). Other variations are exploitative
because they directly augment evolution with hill-climbing: for example, Hybrid Evolutionary
and Hill-Climbing Algorithms and a related method called Scatter Search with Path Relinking.
We discuss all these next.
3.3.1 Elitism
Elitism is simple: we augment the Genetic Algorithm to directly inject into the next population the
fittest individual or individuals from the previous population.32 These individuals are called the
elites. By keeping the best individual (or individuals) around in future populations, this algorithm
begins to resemble (µ + λ), and has similar exploitation properties. This exploitation can cause
premature convergence if not kept in check: perhaps by increasing the mutation and crossover
noise, or weakening the selection pressure, or reducing how many elites are being stored.
A minor catch. If you want to maintain a population size of popsize, and you’re doing crossover,
you’ll need to have popsize, minus the number of elites, be divisible by two, as in this algorithm:
Algorithm 33 The Genetic Algorithm with Elitism

1: popsize ← desired population size
2: n ← desired number of elite individuals " popsize − n should be even
3: P ← {}
6: Best ← !
7: repeat
11: Best ← Pi
12: Q ← {the n fittest individuals in P, breaking ties at random}
13: for (popsize − n)/2 times do
17: Q ← Q ∪ {Mutate(Ca ), Mutate(Cb )}
18: P←Q
20: return Best
32 Elitism was coined by Ken De Jong in his thesis (see Footnote 34, p. 46.).
44
Or you can just throw away an extra crossed-over child if it’d put you over the population size,
as is done in The Genetic Algorithm (Tree-style Genetic Programming Pipeline) (Algorithm 3.3.3).
Elitism is very common. For example, most major multiobjective algorithms (Section 7) are
strongly elitist. Many recent Ant Colony Optimization algorithms (ACO, Section 8.3) are also elitist.
And of course anything resembling (µ + λ), including Scatter Search (Section 3.3.5) is heavily elitist.
Even Particle Swarm Optimization (PSO, Section 3.6) has a kind of elitism in its own regard.
3.3.2 The Steady-State Genetic Algorithm

An alternative to a traditional generational approach to the Genetic Algorithm is to use a steady-
state approach, updating the population in a piecemeal fashion rather than all at one time. This
approach was popularized by the Darrell Whitley and Joan Kauth’s G ENITOR system. The idea
is to iteratively breed a new child or two, assess their fitness, and then reintroduce them directly
into the population itself, killing off some preexisting individuals to make room for them. Here’s a
version which uses crossover and generates two children at a time:
Algorithm 34 The Steady-State Genetic Algorithm

2: P ← {}
5: Best ← !
9: Best ← Pi
10: repeat
11: Parent Pa ← SelectWithReplacement(P) " We first breed two children Ca and Cb
14: Ca ← Mutate(Ca )
15: Cb ← Mutate(Cb )
16: AssessFitness(Ca ) " We next assess the fitness of Ca and Cb
17: if Fitness(Ca ) > Fitness(Best) then
18: Best ← Ca
19: AssessFitness(Cb )
20: if Fitness(Cb ) > Fitness(Best) then
21: Best ← Cb
22: Individual Pd ← SelectForDeath(P)
23: Individual Pe ← SelectForDeath(P) " Pd must be 4= Pe
24: P ← P − { Pd , Pe } " We then delete Pd and Pe from the population
25: P ← P ∪ {Ca , Cb } " Finally we add Ca and Cb to the population
27: return Best
45
The Steady-State Genetic Algorithm has two important features. First, it uses half the memory
of a traditional genetic algorithm because there is only one population at a time (no Q, only P).
Second, it is fairly exploitative compared to a generational approach: the parents stay around in
the population, potentially for a very long time, and thus, like µ + λ and Elitism, this runs the risk
of causing the system to prematurely converge to largely copies of a few highly fit individuals.
This may be exaggerated by how we decide to SelectForDeath. If we tend to select unfit individuals
for death (using, for example, a Tournament Selection based on the least fit in the tournament),
this can push diversity out of the population even faster. More commonly, we might simply select
individuals at random for death. Thus the fit culprits in premature convergence can eventually be
shoved out of the population.33 If we want less exploitation, we may do the standard tricks: use a
relatively unselective operator for SelectWithReplacement, and make Crossover and Mutate noisy.
We could of course generalize this algorithm to replace not just two individuals but some n
individuals all at once. Methods using large values of n (perhaps 50% of the total population size
or more) are often known as Generation Gap Algorithms,34 after Ken De Jong. As n approaches
100%, we get closer and closer to a plain generational algorithm.
3.3.3 The Tree-Style Genetic Programming Pipeline

Genetic Programming (discussed in Section 4.3) is a community interested in using metaheuristics
to find highly fit computer programs. The most common form of Genetic Programming, Tree-
Style Genetic Programming, uses trees as its representation. When doing Tree-Style Genetic
Programming it’s traditional, but hardly required, to use a variant of The Genetic Algorithm
with a special breeding technique due to John Koza.35 Rather than performing crossover and
then mutation, this algorithm first flips a coin. With 90% probability it selects two parents and
performs only crossover. Otherwise, it selects one parent and directly copies the parent into the
new population. It’s this direct copying which makes this a strongly exploitative variant.
A few items of note. First, there’s no mutation: this is not a global algorithm. However the
peculiar version of crossover used in Tree-Style Genetic Programming is so mutative that in practice
mutation is rarely needed. Second, this algorithm could produce one more child than is needed:
just discard it. Third, traditionally the selection procedure is one that is highly selective: Genetic
Programming usually employs Tournament Selection with a tournament size t = 7. Here we go:
33 An interesting question to ask: assuming we have enough memory, why bother deleting individuals at all?
34 There’s a lot of history here. Early ES work employed the now-disused (µ + 1) evolution strategy, where µ parents
(the population) work together to create one new child (see Footnote 15, p. 31). Ken De Jong did early studies of
generation gap methods in Kenneth De Jong, 1975, An Analysis of the Behaviour of a Class of Genetic Adaptive Systems, Ph.D.
thesis, University of Michigan. G ENITOR later popularized the notion of steady-state algorithms. Darrell Whitley and
Joan Kauth, 1988, G ENITOR: A different genetic algorithm, Technical Report CS-88-101, Colorado State University.
35 John R. Koza, 1992, Genetic Programming: On the Programming of Computers by Means of Natural Selection, MIT Press.
46
Algorithm 35 The Genetic Algorithm (Tree-Style Genetic Programming Pipeline)
2: r ← probability of performing direct reproduction " Usually r = 0.1
3: P ← {}
6: Best ← !
7: repeat
11: Best ← Pi
12: Q ← {}
13: repeat " Here’s where we begin to deviate from The Genetic Algorithm
14: if r ≥ a random number chosen uniformly from 0.0 to 1.0 inclusive then
15: Parent Pi ← SelectWithReplacement(P)
16: Q ← Q ∪ {Copy(Pi )}
17: else
21: Q ← Q ∪ { Ca }
22: if || Q|| < popsize then
23: Q ← Q ∪ {Cb }
24: until || Q|| = popsize " End Deviation
25: P←Q
27: return Best
3.3.4 Hybrid Evolutionary and Hill-Climbing Algorithms

There are many many ways in which we can create hybrids of various metaheuristics algorithms, but
perhaps the most popular approach is a hybrid of evolutionary computation and a local improver
such as hill-climbing.
The EA could go in the inner loop and the hill-climber outside: for example, we could extend
Iterated Local Search (ILS, Section 2.6) to use a population method in its inner loop, rather than a
hill-climber, but retain the “Perturb” hill-climber in the outer loop.
But by far the most common approach is the other way around: augment an EA with some
hill-climbing during the fitness assessment phase to revise each individual as it is being assessed.
The revised individual replaces the original one in the population. Any EA can be so augmented:
below is the abstract EA from Algorithm 17 converted into a Hybrid Algorithm.
47
Algorithm 36 An Abstract Hybrid Evolutionary and Hill-Climbing Algorithm
1: t ← number of iterations to Hill-Climb

3: Best ← !
4: repeat
5: AssessFitness(P)
7: Pi ← Hill-Climb(Pi ) for t iterations " Replace Pi in P
9: Best ← Pi
12: return Best
The length of t, of course, is a knob that adjusts the degree of exploitation in the algorithm. If t
is very long, then we’re doing more hill-climbing and thus more exploiting; whereas if t is very
short, then we’re spending more time in the outer algorithm and thus doing more exploring.
There are many other ways to mix an exploitative (and likely local) algorithm with an explorative
(usually global) algorithm. We’ve already seen one example: Hill-Climbing with Random Restarts
(Algorithm 10), which combines a local searching algorithm (Hill-Climbing) with a global algorithm
(Random Search). Another hybrid: Iterated Local Search (Algorithm 16), places Hill-Climbing inside
another, more explorative Hill-Climber. Indeed, the local-improvement algorithm doesn’t even
have to be a metaheuristic: it could be a machine learning or heuristic algorithm, for example. In
general, the overall family of algorithms that combines some kind of global optimization algorithm
with some kind of local improvement algorithm in some way... is often saddled with an ill-considered
name: Memetic Algorithms.36 Though this term encompasses a fairly broad category of stuff,
the lion’s share of memetic algorithms in the literature have been hybrids of global search (often
evolutionary computation) and hill-climbing: and that’s usually how it’s thought of I think.
Perhaps a better term we might use to describe such algorithms could be “Lamarckian Algo-
rithms”. Jean-Baptiste Lamarck was a French biologist around the time of the American revolution
who proposed an early but mistaken notion of evolution. His idea was that after individuals
improved themselves during their lifetimes, they then passed those traits genetically to their off-
spring. For example, horse-like animals in Africa might strain to reach fruit in trees, stretching their
necks. These slightly longer necks were then passed to their offspring. After several generations
of stretching, behold the giraffe. Similarly, these kinds of hybrid algorithms often work by indi-
viduals improving themselves during fitness assessment and then passing on their improvements
36 In my opinion, Memetic Algorithms have little to do with memes, a Richard Dawkins notion which means ideas that
replicate by causing their recipients to forward them to others. Examples include everything from religions to email
chain letters. The term memetic algorithms was notionally justified because memetic algorithm individuals are improved
locally, just as memes might be “improved” by humans before passing them on. But I think the distinguishing feature of
memes isn’t local improvement: it’s replication, even parasitic replication. Nothing in memetic algorithms gets at this.
Richard Dawkins first coined the term meme in Richard Dawkins, 1976, The Selfish Gene, Oxford University Press.
The term memetic algorithms was coined in Pablo Moscato, 1989, On evolution, search, optimization, genetic algorithms
and martial arts: Towards memetic algorithms, Technical Report 158–79, Caltech Concurrent Computation Program,
California Institute of Technology.
48
to their children. Another plausible name would be a “Baldwin Effect Algorithm”, named after
a more plausible variation of Lamarckianism that has found its place in real evolutionary theory.
Much later on we’ll see another example of a Lamarckian algorithm in SAMUEL, an algorithm for
optimizing policies in Section 10.3 with special local-improvement operators.
Another approach to hybridization is to alternate between two disjoint algorithms. For example,
the Learnable Evolution Model (LEM), discussed later in Section 9.1, alternates between evolution
and a machine-learning classification technique.
3.3.5 Scatter Search

Fred Glover’s Scatter Search with Path Relinking37 combines a hybrid evolutionary and hill-
climbing algorithm, line recombination, (µ + λ), and an explicit procedure to inject some diversity
(exploration) into the mix! Standard Scatter Search with Path Relinking is complex and baroque,
but we can describe a simplified version here. The algorithm combines exploitative mechanisms
(hybrid methods, steady-state evolution) with an explicit attempt to force diversity (and hopefully
exploration) into the system. The algorithm starts with a set of initial seeded individuals provided
by you. Then the algorithm tries to produce a large number of random individuals that are very
different from one another and from the seeds. These, plus the seeds, form the population. Then
we do some hill-climbing on each of the individuals to improve them.
We then do the following loop. First, we truncate the population to a small size consisting
of some very fit individuals and some very diverse individuals (to force diversity). Then we
perform some kind of pairing up and crossover (usually using line recombination) on that smaller
population: in our version, we do line recombination on every pair of individuals in the population,
plus some mutating for good measure. Then we do hill-climbing on these new individuals to
improve them, add them to the population, and repeat the loop.
To do the ProduceDiverseIndividual function and the procedure to determine the most diverse
individuals in Q (line 17), you’ll need a distance measure among individuals: )for example, if two
individuals were real-valued vectors !v and !u, use Euclidian distance, that is, ∑i (vi − ui )2 . These
are often metric distances (discussed later in Niching, Section 6.4). From there you could define
the diversity of an individual as its sum distance from everyone else, that is for Population P, the
diversity of Pi is ∑ j distance( Pi , Pj ).
Now we have a way to select based on who’s the “most diverse”. But producing a “diverse”
individual is mostly ad-hoc: I’d generate a lot of individuals, then select a subset of them using a
tournament selection based on maximum diversity from the seeds. Or you could find gene values
uncommon among the seeds and build an individual with them. The simplified algorithm:
37 Glover also invented Tabu Search (Section 2.5). And coined the term “metaheuristics”. It’s tough to pin down the
first papers in Scatter Search. But a good later tutorial is Fred Glover, Manuel Laguna, and Rafael Martı́, 2003, Scatter
search, in Ashish Ghosh and Shigeyoshi Tsutsui, editors, Advances in Evolutionary Computing: Theory and Applications,
pages 519–538, Springer. Glover also attempted a full, detailed template of the process in Fred Glover, 1998, A template
for scatter search and path relinking, in Proceedings of the Third European Conference on Artificial Evolution, pages 1–51,
Springer. The algorithm shown here is approximately derived from these papers.
49
Algorithm 37 A Simplified Scatter Search with Path Relinking
1: Seeds ← initial collection of individuals, defined by you
2: initsize ← initial sample size " The size of the initial population before truncation
3: t ← number of iterations to Hill-Climb
4: n ← number of individuals to be selected based on fitness
5: m ← number of individuals to be selected based on diversity
6: P ← Seeds
7: for initsize − ||Seeds|| times do
8: P ← P ∪ {ProduceDiverseIndividual(P)} " Make an individual very different from what’s in P
9: Best ← !
10: for each individual Pi ∈ P do " Do some hill-climbing
11: Pi ← Hill-Climb(Pi ) for t iterations " Replace Pi in P
14: Best ← Pi
15: repeat " The main loop
16: B ← the fittest n individuals in P
17: D ← the most diverse m individuals in P " Those as far from others in the space as possible
18: P ← B∪D
19: Q ← {}
21: for each individual Pj ∈ P where j 4= i do
22: Children Ca , Cb ← Crossover(Copy(Pi ), Copy(Pj )) " Line Recombination, Algorithm 28
23: Ca ← Mutate(Ca ) " Scatter Search wouldn’t do this normally: but I would
24: Cb ← Mutate(Cb ) " Likewise
25: Ca ← Hill-Climb(Ca ) for t iterations
26: Cb ← Hill-Climb(Cb ) for t iterations
27: AssessFitness(Ca ) " We next assess the fitness of Ca and Cb
29: Best ← Ca
32: Best ← Cb
33: Q ← Q ∪ {Ca , Cb }
34: P ← Q∪P
36: return Best
50
3.4 Differential Evolution: an Adaptive Mutation Algorithm
Differential Evolution (DE) determines the size of Mutates largely based on the current variance in
the population. If the population is spread out, Mutate will make major changes. If the population
is condensed in a certain region, Mutates will be small. It’s an adaptive mutation algorithm (like
the one-fifth rule in Evolution Strategies). DE was developed by Kenneth Price and Rainer Storn.38
DE’s mutation operators employ vector addition and subtraction,
so it really only works in metric vector spaces (booleans, metric integer
spaces, reals). DE has a variety of mutation operators, but the early
one described here is common and easy to describe. For each member
i of the population, we generate a new child by picking three indi-
viduals from the population and performing some vector additions B C
and subtractions among them. The idea is to mutate away from one
of the three individuals (!a ) by adding a vector to it. This vector is
Child A
created from the difference between the other two individuals !b −!c.
If the population is spread out, !b and !c are likely to be far from one
another and this mutation vector is large, else it is small. This way, if Figure 15 Differential Evolu-
the population is spread throughout the space, mutations will be much tion’s primary mutation opera-
bigger than when the algorithm has later converged on fit regions of tor. A copy of individual A is
mutated by adding to it the vec-
the space. The child is then crossed over with !i. (Differential Evolution tor between two other individu-
has lots of other mutation variations not shown here). als B and C, producing a child.
Finally, after we have built up a new group of children, we compare
each child with the parent which created it (each parent created a single child). If the child is better
than the parent, it replaces the parent in the original population.
The new locations of children are entirely based on the existing parents and which combinations
we can make of adding and subtracting them. This means that this algorithm isn’t global in the
sense that any point in the space is possible: though through successive choices of individuals, and
mutating them, we can hone in on certain spots in the space. Also oddly this algorithm traditionally
mutates each individual in turn. Perhaps better would be either to mutate all of them in parallel (in
a generational fashion) or to pick i at random each time (steady-state style).
It”s crucial to note that Differential Evolution “selects” individuals in a way quite different from
what we’ve seen so far. A child is created by mutating existing individuals largely picked at random
from the population. So where’s the selection? It comes after generating a child, when it competes
for survival with a specific individual already in the population. If the child is fitter, it replaces that
individual, else the child is thrown away. This hill-climbing-ish approach to selection is a variation
of survival selection (as opposed to parent selection).39
Below we show one simple implementation of Differential Evolution, as described above.
Note that in this code we will treat the population as a vector, not a collection: this is to make
the pseudocode a bit more clear. Also, note that since Differential Evolution always uses vector
representations for individuals, we’ll treat individuals both as individuals (such as Qi and as
vectors (such as !a) interchangeably. Here we go:
38 DE grew out of a series of papers as it evolved, but one of its better known papers, if not the earliest, is Rainer Storn
and Kenneth Price, 1997, Differential evolution: A simple and efficient heuristic for global optimization over continuous
spaces, Journal of Global Optimization, 11(4), 341–359. Price, Storn, and Jouni Lampinen later wrote a pretty big book on
the subject: Kenneth Price, Rainer Storn, and Journi Lampinen, 2005, Differential Evolution: A Practical Approach to Global
Optimization, Springer.
39 See footnote 14.
51
Algorithm 38 Differential Evolution (DE)
1: α ← mutation rate " Commonly between 0.5 and 1.0, higher is more explorative
3: P←$% " Empty population (it’s convenient here to treat it as a vector), of length popsize
4: Q←! " The parents. Each parent Qi was responsible for creating the child Pi
5: for i from 1 to popsize do
6: Pi ← new random individual
7: Best ← !
8: repeat
11: if Q 4= ! and Fitness(Qi ) > Fitness(Pi ) then
12: Pi ← Qi " Retain the parent, throw away the kid
14: Best ← Pi
15: Q←P
16: for each individual Qi ∈ Q do " We treat individuals as vectors below
17: !a ← a copy of an individual other than Qi , chosen at random with replacement from Q
18: !b ← a copy of an individual other than Qi or !a, chosen at random with replacement from Q
19: !c ← a copy of an individual other than Qi , !a, or !b, chosen at random with replacement from Q
20: d! ← !a + α(!b −!c) " Mutation is just vector arithmetic
21: Pi ← one child from Crossover(d,! Copy(Qi ))
22: until Best is the ideal solution or we ran out of time
23: return Best
Crossover can be anything: but one common approach is to do a uniform crossover (Algorithm
25), but guarantee that at least one gene from Qi (the gene is chosen at random) survives in Pi .
3.5 Variations on the Theme of Exploitation versus Exploration

The evolutionary computation algorithms presented are mostly variations on the same theme: the
manipulation of resampling to do various kinds of exploration and exploitation. There are various
ways to do this, but overall, here are the big five parameters we can set to explore more or exploit
more in different ways:
• The size of the population A very large population approaches random search. A very
small population approaches hill-climbing.
• How likely a fit parent is chosen over an unfit parent (Selection Pressure) High selection
pressure approaches hill-climbing. Low selection pressure approaches a random walk (note:
not random search).
• How many children are generated from a parent Many children samples a lot near parents,
similar to steepest-ascent hill-climbing. Few children is similar to plain hill-climbing.
52
• How different children are from their parents (Mutation Rate) A high mutation rate
fuzzes out the samples more, approaching random search. Very small mutation rates finesse
local optima more precisely.
• Whether parents can stick around (Elitism or Survival Selection) If parents cannot stick
around, the algorithm is more like a random walk. Else it tends to exploit the parents’ optima.
In cases like DE, a child must defeat its parent to even be included in the population.
3.6 Particle Swarm Optimization

Particle Swarm Optimization (PSO) is a stochastic optimization technique somewhat similar
to evolutionary algorithms but different in an important way. It’s modeled not after evolution
per se, but after swarming and flocking behaviors in animals. Unlike other population-based
methods, PSO does not resample populations to produce new ones: it has no selection of any kind.
Instead, PSO maintains a single static population whose members are Tweaked in response to new
discoveries about the space. The method is essentially a form of directed mutation. The technique
was developed by James Kennedy and Russell Eberhart in the mid-1990s.40
Like Differential Evolution, PSO operates almost exclusively in multidimensional metric, and
usually real-valued, spaces. This is because PSO’s candidate solutions are Mutated towards the best
discovered solutions so far, which really necessitates a metric space (it’s nontrivial to Mutate, say, a
tree “towards” another tree in a formal, rigorous fashion).
Because of its use in real-valued spaces, and because PSO is inspired by flocks and swarms,
PSO practitioners tend to refer to candidate solutions not as a population of individuals but as a
swarm of particles. These particles never die (there is no selection). Instead, the directed mutation
moves the particles about in the space. A particle consists of two parts:
• The particle’s location in space, !x = $ x1 , x2 , ...%. This is the equivalent, in evolutionary

algorithms, of the individual’s genotype.
• The particle’s velocity, !v = $v1 , v2 , ...%. This is the speed and direction at which the particle is
traveling each timestep. Put another way, if !x (t−1) and !x (t) are the locations in space of the
particle at times t − 1 and t respectively, then at time t, !v = !x (t) − !x (t−1) .
Each particle starts at a random location and with a random velocity vector, often computed by
choosing two random points in the space and using half the vector from one to the other (other
options are a small random vector or a zero vector). We must also keep track of a few other things:
• The fittest known location !x ∗ that !x has discovered so far.
• The fittest known location !x + that any of the informants of !x have discovered so far. In early
versions of the algorithm, particles were assigned “grid neighbors” which would inform
them about known best-so-far locations. Nowadays the informants of !x are commonly a
small set of particles chosen randomly each iteration. !x is always one of its own informants.
• The fittest known location !x ! that has been discovered by anyone so far.
40 Among the earliest papers on PSO is James Kennedy and Russell Eberhart, 1995, Particle swarm optimization, in
Proceedings of IEEE International Conference on Neural Networks, pages 1942–1948. Eberhart, Kennedy, and Yuhui Shi later
wrote a book on the topic: James Kennedy, Russell Eberhart, and Yuhui Shi, 2001, Swarm Intelligence, Morgan Kaufmann.
53
Each timestep we perform the following operations:
1. Assess the fitness of each particle and update the best-discovered locations if necessary.
2. Determine how to Mutate. For each particle !x, we update its velocity vector !v by adding in,
to some degree, a vector pointing towards !x ∗ , a vector pointing towards !x + , and a vector
pointing towards !x ! . These are augmented by a bit of random noise (different random values
for each dimension).
3. Mutate each particle by moving it along its velocity vector.
The algorithm looks like this:
Algorithm 39 Particle Swarm Optimization (PSO)

1: swarmsize ← desired swarm size
2: α ← proportion of velocity to be retained
3: β ← proportion of personal best to be retained
4: γ ← proportion of the informants’ best to be retained
5: δ ← proportion of global best to be retained
6: % ← jump size of a particle
7: P ← {}
8: for swarmsize times do
9: P ← P ∪ {new random particle !x with a random initial velocity !v}
−→
10: Best ← !
11: repeat
12: for each particle !x ∈ P with velocity !v do
13: AssessFitness(!x)
−→ −→
14: if Best = ! or Fitness(!x) > Fitness(Best) then
−→
15: Best ← !x
16: for each particle !x ∈ P with velocity !v do " Determine how to Mutate
17: ∗
!x ← previous fittest location of !x
18: !x + ← previous fittest location of informants of !x " (including !x itself)
19: !x ! ← previous fittest location any particle
20: for each dimension i do
21: b ← random number from 0.0 to β inclusive
22: c ← random number from 0.0 to γ inclusive
23: d ← random number from 0.0 to δ inclusive
24: vi ← αvi + b( xi∗ − xi ) + c( xi+ − xi ) + d( xi! − xi )
25: for each particle !x ∈ P with velocity !v do " Mutate
26: !x ← !x + %!v
−→
−→
28: return Best
54
This implementation of the algorithm relies on five parameters:
• α: how much of the original velocity is retained.
• β: how much of the personal best is mixed in. If β is large, particles tend to move more
towards their own personal bests rather than towards global bests. This breaks the swarm
into a lot of separate hill-climbers rather than a joint searcher.
• γ: how much of the informants’ best is mixed in. The effect here may be a mid-ground
between β and δ. The number of informants is also a factor (assuming they’re picked at
random): more informants is more like the global best and less like the particle’s local best.
• δ: how much of the global best is mixed in. If δ is large, particles tend to move more towards
the best known region. This converts the algorithm into one large hill-climber rather than
separate hill-climbers. Perhaps because this threatens to make the system highly exploitative,
δ is often set to 0 in modern implementations.
• %: how fast the particle moves. If % is large, the particles make big jumps towards the
better areas — and can jump over them by accident. Thus a big % allows the system to move
quickly to best-known regions, but makes it hard to do fine-grained optimization. Just like in
hill-climbing. Most commonly, % is set to 1.
Consider to what degree these five parameters map to the various Variations on Exploration
and Exploitation discussed in Section 3.5.
55
56
4 Representation
Most techniques discussed later are typically done with population-based algorithms. So from now on we will usually
use Evolutionary Computation versions of terms: individual instead of candidate solution; fitness instead of quality, etc.
The representation of an individual is the approach you take to constructing, tweaking, and
presenting the individual for fitness assessment. Although often we’ll refer to the representation
as the data structure used to define the individual (a vector, a tree, etc.) it’s useful to think of the
representation not as the data type but instead simply as two functions:
• The initialization function used to generate a random individual.
• The Tweak function, which takes one individual (or more) and slightly modifies it.
To this we might add...
• The fitness assessment function.
• The Copy function.
These functions are the only places where many optimization algorithms deal with the internals
of individuals. Otherwise the algorithms treat individuals as black boxes. By handling these
functions specially, we can separate the entire concept of representation from the system.
Much of the success or failure of a metaheuristic lies in the design of the representation of
the individuals, because their representation, and particularly how they Tweak, has such a strong
impact on the trajectory of the optimization procedure as it marches through the fitness landscape
(that is, the quality function). A lot of the black magic involved in constructing an appropriate
representation lies in finding one which improves (or at least doesn’t worsen) the smoothness
of the landscape. As mentioned earlier, the smoothness criterion was approximately defined as:
individuals which are similar to each other tend to behave similarly (and thus tend to have similar
fitness), whereas individuals dissimilar from one another make no such promise.
The smoother a landscape, the fewer hills
Unimodal Needle in a Haystack
it has and the more it begins to resemble a uni-
modal landscape, as shown in Figure 16. Re-
call that this isn’t a sufficient criterion though,
as needle-in-a-haystack or (worse) deceptive
environments are highly smooth, yet can be
extremely challenging for an optimization al-
gorithm. Noisy
(or “Hilly” or “Rocky”)
Deceptive
When we refer to individuals being simi-

lar, we mean that they have similar genotypes,
and when we refer to individuals as behaving
similarly, we mean that they have similar phe-
notypes.41 What do we mean by similar geno-
types? Generally genotype A is similar to geno-
Figure 16 Four fitness landscapes. Repeats Figure 6.
type B if the probability is high that Tweaking
41 Recall that, in evolutionary computation at least, the phrase “genotype” refers to how the individual appears to the
genetic operators (perhaps it’s a vector, or a tree), and the phrase “phenotype” refers to how (not how well) the individual
performs when evaluated for fitness assessment.
57
A will result in B (or vice versa). Thus things are similar not because their genotypes look similar,
but because they are near each other in the space with respect to your choice of the Tweak operation.
It’s tempting to think of a stochastic optimization system as largely working in genotype space,
then translating the genotypes to phenotypes for purposes of evaluation. But when thinking about
the effect of representations, it’s better to consider the other way around: an individual’s natural
arrangement is its phenotype, and when the algorithm needs to make a new individual, it translates
the phenotype to a genotype, Tweaks it, then translates back to the phenotype. Commonly we refer
to phenotype→genotype translation as encoding, and the reverse as decoding. Thus we can think
of this process as:
Parent Phenotype −→ Encode −→ Tweak −→ Decode −→ Child Phenotype
This view helps us see the perils of poor encoding choices. Imagine that your individuals take
the phenotypical form, for some reason, of Rubik’s Cube configurations. You’d like that Tweak
operator to make small changes like rotating a side, etc. If you used a genotype in the form of a
Rubik’s Cube, you’re all set: the Tweak operator already does exactly what you want. But imagine
if your encoding operation was as follows:
Parent −→ Do 20 specific unusual moves −→ Tweak −→ Undo those 20 moves −→ Child
You can imagine that after doing the twenty moves, a single twist of one side (the Tweak) will
have huge consequences after you undo those twenty moves. It causes almost total randomization
from parent to child. Lesson: you want an encoding/decoding mechanism which doesn’t cause
your carefully-selected, smooth Tweak operations to cause the phenotype space to go haywire.
This isn’t just of academic concern. In the past, Genetic Algorithm folks used to encode
everything as a binary vector of fixed length. The reasoning was: if there’s only one genotype, we
could develop a canonical Genetic Algorithm as a library function, and the only differences of
significance would be the encoding procedure.
As it turns out, this wasn’t all that good of an Phenotype Genotype Gray Code Fitness
idea. Consider the situation where an individual 0 0000 0000 0
consists of a single integer from 0 to 15. We’d rep- 1 0001 0001 1
resent it as a vector of 4 bits. The fitness function 2 0010 0011 2
is shown at right. Notice that it increases until 3 0011 0010 3
8, and then “falls off the cliff” at 9. This fitness 4 0100 0110 4
function abuses a bad feature in the genotype: 5 0101 0111 5
what is known in the Genetic Algorithm com- 6 0110 0101 6
munity as a Hamming cliff, located at the jump 7 0111 0100 7
from 7 to 8. A Hamming cliff is where, to make 8 1000 1100 8
a small change in the phenotype or fitness, you 9 1001 1101 0
must make a very large change in the genotype. 10 1010 1111 0
For example, to mutate 7 (0111) into 8 (1000), 11 1011 1110 0
you have to make four bit-flips in succession. 12 1100 1010 0
The function at right is hard to optimize because 13 1101 1011 0
to get to 8, notionally you could approach from 14 1110 1001 0
7 (requiring four lucky mutations) or you could 15 1111 1000 0
approach from 9 or 10 (which aren’t often going
to be selected, because of bad fitness). Table 2 A fitness function that exploits a Hamming Cliff.
58
Now consider instead representing the individual not by the binary encoding genotype shown
above but rather its Gray code42 encoding shown next to it. This encoding has an interesting
property: each successive number differs from its previous number by only one bit flip. And 15
differs from 0 by only one bit flip. Thus if we’re at 7 (Gray code 0100) we can easily mutate to 8
(Gray code 1100). Hamming cliff problem solved. By the way, Gray-coding is easy to do:
Algorithm 40 A Gray Coding

v ← boolean vector encoding a standard binary number $v1 , v2 , ...vl % to be converted to Gray code
1: !
2: w
! ← Copy(!v)
4: if vi−1 is true then
5: wi ← ¬(vi )
6: return w
!
The point of this exercise is not to convince you to use Gray codes: indeed, we can construct
nasty fitness functions which cause problems for Gray codes as well, and Gray coding is somewhat
old fashioned now. The point is to illustrate the notion of smoothness and its value. If you encode
your individual such that small changes in the genotype (like one bit flip) are somewhat more
likely to result in small changes in the fitness, you can help your optimizer.
One heuristic approach to smooth fitness landscapes is to make the genotype as similar to the
phenotype as possible: if your phenotype is a graph structure, let the genotype be a graph structure
as well. That way your fitness function may still be hilly but at least you’re not making it even
hillier by running it through an unfortunate encoding. But remember that this is thinking of the
representation as if it’s a data structure, when it’s not. It’s largely two functions: the initialization
function and theTweak function.
Much of Representation Is an Art, Not a Science How are you going to Tweak a graph structure
in a smooth way? No, seriously. Certain representations (notably fixed-length vectors of booleans
or of floating-point values) are very well understood and there’s a bunch of good theory around
them. But many representations are still basically ad-hoc. Many of the algorithms and ideas in this
section should not be taken as directions, or even recommendations, but suggestions of one particular
possible way to do representations that maintain smoothness properties. We’ll first take care of the
easy, well-understood one that we’ve seen before a lot: vectors.
4.1 Vectors
Just to be clear, by vectors we mean fixed-length one-dimensional arrays. We’ll get to arbitrary-
length lists in Section 4.4. Vectors usually come in three flavors: boolean, real-valued, and integer.43
The first two — boolean and real-valued vectors — we’ve seen a lot so far. As a result we’ve built
up several initialization, mutation, and crossover algorithms for them. In summary:
42 AfterFrank Gray, who developed it in 1947 at Bell Labs to reduce errors in the output of phone system switches.
43 There’s no reason you couldn’t have a vector of trees, or a vector of rules, or a vector where some elements were
reals and others were booleans, etc. (In fact, we’ll see vectors of trees and rules later on in Section 4.3.4!) You just need to
be more careful with your mutation and initialization mechanisms.
59
Boolean Vectors
Initialization
Generate a Random Bit-Vector Algorithm 21 Page 35
Mutation
Bit-Flip Mutation Algorithm 22 Page 36
Floating-Point Vectors
Initialization
Generate a Random Real-Valued Vector Algorithm 7 Page 17
Mutation
Bounded Uniform Convolution Algorithm 8 Page 17
Gaussian Convolution Algorithm 11 Page 21
Floating-Point-Specific Crossover
Line Recombination Algorithm 28 Page 40
Intermediate Recombination Algorithm 29 Page 40
Vector Crossover (applies to any vector type)

One-Point Crossover Algorithm 23 Page 36
Two-Point Crossover Algorithm 24 Page 37
Uniform Crossover Algorithm 25 Page 37
Uniform Crossover among K Vectors Algorithm 27 Page 39
Integer Vectors We’ve not seen integer vectors yet: and integer vectors have a twist to consider.
What do the integers in your vector represent? Do they define a set of unordered objects (1=China,
2=England, 3=France, ...) or do they form a metric space (IQ scores, or street addresses, or final
course grades) where the distance between, say, 4 and 5 is greater than the distance between 1 and
5? Mutation decisions often center on whether the space is a metric space.
The remainder of this section will focus on integer vectors, but it also gives some discussion
relevant to initialization and mutation of all vector types.
4.1.1 Initialization and Bias

Creating random initial vectors is usually just a matter of picking
each vector position vi uniformly among all possible values. If you
have some knowledge about your problem, however, you could
bias the system by tending to pick values in certain regions of the
space. For example, if you believe that better solutions usually lie
in the regions where v1 = v2 × v3 , you could emphasize generating
vectors in those regions.
Another way to bias the initial configuration of your population
is to seed the initial population with pre-chosen individuals of your
Figure 17 Our bipedal robot.
own design. For example, my students were trying to optimize
vectors which defined how a bipedal robot walked, kicked, etc.
These vectors translated into joint angles and movements for the many motors on the robot. Rather
than start with random values, the vast majority of which were nonsense, they instead chose to
wire a student up to a 3D tracker and have him perform the motions. They then converted the
resulting data into joint angle movements, which they used to seed the initial population.
60
Some suggestions. First, biasing is dangerous. You may think you know where the best solutions
are, but you probably don’t. So if you bias the initial configuration, you may actually make it
harder for the system to find the right answer. Know what you’re getting into. Second, even if you
choose to bias the system, it may be wise to start with values that aren’t all or exactly based on your
heuristic bias. Diversity is useful, particularly early on.
4.1.2 Mutation
It’s rare that you’d mutate floating-point vectors with anything other than Guassian convolution
(or some similar distribution-based noise procedure). Likewise, bit-vectors are typically mutated
using bit-flip mutation. For integer vectors, it depends. If your representation treats integers as
members of a set, the best you may be able to do is randomize each slot with a given probability:
Algorithm 41 Integer Randomization Mutation

v ← integer vector $v1 , v2 , ...vl % to be mutated
1: !
2: p ← probability of randomizing an integer " Perhaps you might set p to 1/l or lower
5: vi ← new random legal integer
6: return ! v
If instead your integers represent a metric space, you might wish to mutate them in a manner
similar to gaussian convolution, so that the changes to integers tends to be small. One of a great
many ways to do this is to keep flipping a coin until it comes up heads, and do a random walk of
that length.44 This creates noise centered around the original value, and is global.
Algorithm 42 Random Walk Mutation

v ← integer vector $v1 , v2 , ...vl % to be mutated
1: !
2: p ← probability of randomizing an integer " Perhaps you might set p to 1/l or lower
3: b ← coin-flip probability " Make b bigger if you have many legal integer
values so the random walks are longer
6: repeat
7: n ← either a 1 or -1, chosen at random.
8: if vi + n is within bounds for legal integer values then
9: vi ← vi + n
10: else
11: vi ← vi − n
12: until b < random number chosen uniformly from 0.0 to 1.0 inclusive
13: return !v
44 Note: I just made up this mutator, but it’s probably not bad. And someone else probably already invented it.
61
Point Mutation The mutation methods discussed so far all have the same property: every gene
in the genome has an independent probability of being mutated. Perhaps you may have thought
of a different approach: pick a single random gene, then mutate that gene, and you’re done. (Or
perhaps pick n genes at random and mutate them). Such point mutation methods are sometimes
useful but are often dangerous.
First the useful part: there exist some problems where you can make progress through the
space by changing a single gene, but if you change several genes at a time, even by a small amount,
it’s tougher to make progress. The Mona Lisa picture on the front page is an example of this: the
genome consists of some m polygons with random colors. Change one polygon at a time, by a fair
bit, and you can eventually eek out a Mona Lisa. Change n polygons (or even all m polygons) at
one time, even through small perturbation, and it turns out to be quite difficult to get a better child.
But beware: it’s very easy to construct problems where point mutation x
is quite bad indeed. Consider simple boolean individuals of the form 0 1
$ x, y%, where x and y can each be 1 or 0, and we’re doing a simple hill- 0 5 -100
climber (or (1 + 1) if you will). The problem uses the fitness function y 1 -100 10
shown in Table 3, and our intrepid initial candidate solution starts at
$0, 0%, which at present has a fitness of 5. Our mutation function flips a Table 3 A trivial boolean fit-
single gene. If we flipped gene x, we’d wind up in $1, 0%, with a fitness of ness function which is hos-
tile to point mutation.
-100, which would get promptly rejected. On the other hand, if we flipped
gene y, we’d wind up in $0, 1%, also with a fitness of -100. There’s no way
to get to the optimum $1, 1% without flipping both genes at the same time. But our mutation operator
won’t allow that. The issue is that point mutation is not a global operator: it can only make horizontal
moves through the space, and so cannot reach all possible points in one jump. In summary: point
mutation can sometimes be useful, but know what you’re getting into.
4.1.3 Recombination
So far we’ve seen three kinds of general-purpose vector recombination: One- and Two-point
Crossover, and Uniform Crossover. Additionally we’ve seen two kinds of recombination designed
for real-valued number recombination: Line Recombination and Intermediate Recombination.
Of course you could do a similar thing as these last two algorithms with metric-space integers:
Algorithm 43 Line Recombination for Integers

1: !
2: w
3: p ← positive value which determines how far along the line a child can be located

7: repeat
8: t ← αvi + (1 − α)wi
9: s ← βwi + (1 − β)vi
10: until 0t + 1/21 and 0s + 1/21 are within bounds " The 0... + 1/21 bit is rounding
11: vi ← 0t + 1/21
12: wi ← 0s + 1/21
62
Algorithm 44 Intermediate Recombination for Integers
1: !
2: w
3: p ← positive value which determines how far long the line a child can be located
5: repeat
8: t ← αvi + (1 − α)wi
9: s ← βwi + (1 − β)vi
10: until 0t + 1/21 and 0s + 1/21 are within bounds
11: vi ← 0t + 1/21
12: wi ← 0s + 1/21
Phenotype-Specific Mutation or Crossover Last but not least, you might try instead to perform
mutation or crossover on your representations in a manner that makes sense with regard to their
phenotype. For example, what if your phenotype is a matrix, and you’re using vectors to represent
those matrices? Perhaps your recombination operators should take into consideration the two-
dimensional nature of your phenotype. You might design a recombination operator which does
two one-point crossovers to slice out a rectangular region:
     
1 4 7 21 99 46 1 4 46
 9 2 3  crossed over with  31 42 84  −→  9 2 84 
8 5 6 23 67 98 23 67 98
This leads us to using representations more apropos to your problem: so on to more complex
representations. Remember all that talk about the value of smoothness? Hold onto your hat because
when you get to nastier representations, guaranteeing smoothness becomes very hard indeed.
4.2 Direct Encoded Graphs

Graphs are just about the most complex of various representations, but it’s useful to discuss them
next. Why would you want to come up with an optimal graph structure? Graphs are used
to represent many things: neural networks, finite-state automata or Petri nets or other simple
computational devices, electrical circuits, relationships among people, etc. Correspondingly, there
are lots of kinds of graph structures, such as directed graphs, undirected graphs, graphs with labels
on the edges or nodes, graphs with weights (numbers) on the edges rather than labels, recurrent
graphs, feed-forward (non-recurrent) graphs, sparse or dense graphs, planar graphs, etc. It depends
on your problem. A lot of the decisions with regard to Tweaking must work within the constraints
of the graph structure you’ve decided on.
First note that you don’t need a special representation if your graph structure is fixed and
you’re just finding weights or labels. For example, if you’re developing a neural network with a
63
fixed collection of edges, there’s no need to discover the structure of this network (it’s fixed!). Just
discover the weights of the edges. If you have 100 edges, just optimize a vector of 100 real-valued
numbers, one per edge weight, and you’re done. Thus most “graph representations” of interest here
are really arbitrary-structured graph representations. Such structures have been around for a very
long time. Larry Fogel developed Evolutionary Programming, probably the earliest evolutionary
algorithm, specifically to discover graph structures in the form of finite-state automata.45
There are generally two approaches to developing graph structures (and certain other complex
structures): direct encoding and indirect (or developmental) encoding. Direct encoding stores
the exact edge-for-edge, node-for-node description of the graph structure in the representation
itself. Indirect encoding has the representation define a small program or set of rules of some kind
which, when executed, “grow” a graph structure.
Why would you do an indirect encoding? Perhaps when you wish to cross over certain traits in
your graph structure described by subsets of those rules which are bundled together. Or perhaps
if your rules recursively cause other rules to fire, you may view certain sets of rules as functions
or modules which always produce the same subgraph. Thus if your optimal graph structures are
highly repetitive, you can take advantage of this by evolving a single function which produces that
repetitive element rather than having to rediscover the subgraph over and over again during the
search process. If the graph has little repetition in it (for example, neural network weights tend to
have little repetition among them) and is very dense, a direct encoding might be a better choice.
Because indirect encodings represent the graph in a non-graph way (as a tree, or a set of rules, or a
list of instructions to build the graph, etc.), we’ll discuss them later (in Sections 4.3.6 and 4.5). For
now, we consider direct encodings.
The simplest direct encoding is a full adjacency matrix. Here we have settled on an absolute
maximum size for our graph. Let’s say we need to create a recurrent directed graph structure and
have decided that our graph will contain no more than 5 nodes and have no more than one edge
between any two nodes. Let’s also say that self-edges are allowed, and we need to find weights for
the edges. We could simply represent the graph structure as a 5 × 5 adjacency matrix describing
the edges from every node to every other node:
 
0.5 0.7 −0.1 0.2 Off
 Off −0.5 −0.8 0.4 Off 
 
 0.6 0.7 0.8 Off −0.4 
 
−0.1 Off Off 0.2 Off 
0.2 Off −0.7 Off Off
“Off” in position $i, j% means “there is no edge connecting j to i”. If we want fewer than 5
nodes, we could just assign all the weights going in or out of a node to be “Off”. We could represent
this matrix in many ways. Here are two. First, we might have a single vector of length 25 which
stores all the weights, with “Off” being represented as 0.0. Or we could represent the matrix as two
vectors, a real-valued one which stores all the weights, and a boolean one which stores whether or
not an edge is “On” or “Off”. Either way, we could use standard crossover and mutation operators,
though we might want to be careful about changing “Off” values. If we used the two-vector
45 For Fogel’s thesis, in which these ideas were advanced, see Footnote 18, p. 34.
64
version, that’s done for us for free. If we just use a single real-valued vector, we could create a
modified Gaussian Convolution algorithm which only sometimes turns edges on or off:
Algorithm 45 Gaussian Convolution Respecting Zeros

1: !
2: p ← probability of changing an edge from “On” to “Off” or vice versa
3: σ2 ← variance of gaussian distribution to convolve with
8: if vi = 0.0 then " Turn “On”: pick a random edge weighting
9: vi ← random number chosen uniformly from 0.0 to 1.0 inclusive
10: else " Turn “Off”
11: vi ← 0.0
12: else if vi 4= 0.0 then " Mutate an existing “On” weight
13: repeat
14: n ← random number chosen from the Normal distribution N (0, σ2 ) " See Algorithm 12
16: vi ← vi + n
17: return !v
The disadvantage of this approach is that once an edge is turned “Off”, when it’s turned back
“On”, its previously carefully-optimized weight is lost. Perhaps the two-vector approach might
yield better results.
If we don’t have a maximum size for our graph, we might need to use an arbitrary directed
graph structure, an approach done very early on (in EP) but popularized by Peter Angeline, Greg
Saunders, and Jordan Pollack’s GNARL.46 Here our representation isn’t a vector: it’s an actual
graph, stored however we like. To do this, we need to create custom initialization and mutation or
crossover operators to add and delete nodes, add and delete edges, relabel nodes and edges, etc.
A similar approach is taken in N EAT,47 Ken Stanley and Risto Miikkulainen’s method for
optimizing feed-forward neural networks. N EAT represents a graph as two sets, one of nodes and
one of vectors. Each node is simply a node number and a declaration of the purpose of the node (in
neural network parlance: an input, output, or hidden unit). Edges are more interesting: each edge
contains, among other things, the nodes the edge connected (by number), the weight of the edge,
and the birthday of the edge: a unique counter value indicating when the edge had been created.
The birthday turns out to be useful in keeping track of which edges should merge during crossover,
as discussed in Section 4.2.3.
46 PeterJ. Angeline, Gregory M. Saunders, and Jordan P. Pollack, 1994, An evolutionary algorithm that constructs
recurrent neural networks, IEEE Transactions on Neural Networks, 5(1), 54–65.
47 Kenneth O. Stanley and Risto Miikkulainen, 2002, Evolving neural networks through augmenting topologies,
Evolutionary Computation, 10(2), 99–127.
65
4.2.1 Initialization
Creating an initial graph structure is mostly informed by the kind of graphs you think you need.
First, we might decide on how many nodes and edges we want. We could pick these from some
distribution — perhaps a uniform distribution from 1 to some large value. Or we might choose a
them from a distribution which heavily favors small numbers, such as the Geometric Distribution.
This distribution is formed by repeatedly flipping a coin with probability p until it comes up heads:
Algorithm 46 Sample from the Geometric Distribution

1: p ← probability of picking a bigger number
2: m ← minimum legal number
3: n ← m−1
4: repeat
5: n ← n+1
6: until p < random number chosen uniformly from 0.0 to 1.0 inclusive
7: return n
The larger the value of p, the larger the value of n on average, using the equation E(n) =
m + p/(1 − p). For example, if m = 0 and p = 3/4, then n will be 3 on average, while if p = 19/20,
then n will be 19 on average. Beware that this distribution has a strong tendency to make lots of
small values. It’s easy to compute, but may wish to use a less skewed distribution.
Once we have our node and edge counts, we can build a graph by laying out the nodes first,
then filling in the edges:
Algorithm 47 Build A Simple Graph

1: n ← chosen number of nodes
2: e ← chosen number of edges
3: f (j, k, Nodes, Edges) ← function which returns ’true’ if an edge from j to k is allowed
4: set of nodes N ← { N1 , ...Nn } " Brand new nodes

5: set of edges E ← {}
6: for each node Ni ∈ N do
7: ProcessNode(Ni ) " Label it, etc., whatever
8: for i from 1 to e do
9: repeat
10: j ← random number chosen uniformly from 1 to n inclusive
11: k ← random number chosen uniformly from 1 to n inclusive
12: until f ( j, k, N, E) returns ’true’
13: g ← new edge from Nj to Nk
14: ProcessEdge(g) " Label it, weight it, undirect it, whatever
15: E ← E ∪ { g}
16: return N, E
66
Note the ProcessNode and ProcessEdge functions, which give you a place to label and weight
edges and nodes. A difficulty with this approach is that we could wind up with a disjoint graph:
you may need to adjust this algorithm to guarantee connectedness. Another very common graph
representation is a directed acyclic graph, where all edges go from later nodes to earlier ones:
Algorithm 48 Build a Simple Directed Acyclic Graph

2: D ( m ) ← probability distribution of the number of edges out of a node, given number of in-nodes m

8: for i from 2 to n do
9: p ← random integer ≥ 1 chosen using D (i − 1)
10: for j from 1 to p do
11: repeat
12: k ← random number chosen uniformly from 1 to i − 1 inclusive
13: until f (i, k, N, E) returns ’true’
14: g ← new edge from Ni to Nk
15: ProcessEdge(g)
16: E ← E ∪ { g}
17: return N, E
This representation is connected but of course there are no loops. Anyway, these algorithms are
only to give you ideas: definitely don’t rely on them! Do it right. There are tons of (much better)
randomized graph-building algorithms: consult any general algorithms text.
4.2.2 Mutation
One of many ways to mutate an arbitrary graph is to pick some number n of mutations, then n
times do any of:
• With α1 probability, delete a random edge.
• With α2 probability, add a random edge (if using N EAT, this edge would get a brand new
birthday number; see Section 4.2.3 next).
• With α3 probability, delete a node and all its edges (yeesh!)
• With α4 probability, add a node
• With α5 probability, relabel a node
• With α6 probability, relabel an edge ... etc. ...
67
... where ∑i αi = 1.0. Obviously some of these operations are very mutative, and thus perhaps
should have a smaller probability. Keep in mind that small, common changes should result in small
fitness changes, that is, more mutative operations should be done less often. Last, how do we pick
a value for n? Perhaps we might pick uniformly between some values 1...M. Or we might choose a
value from the Geometric Distribution again.
4.2.3 Recombination
Crossover in graphs is such a mess that many people don’t do it at all. How do you cross over
graphs in a meaningful way? That is, transferring essential and useful elements from individual to
individual without having crossover basically be randomization?
To cross over nodes and edges we often need to get subsets of such things. To select a subset:
Algorithm 49 Select a Subset

1: S ← original set
2: p ← probability of being a member of the subset
3: subset S) ← {}
4: for each element Si ∈ S do
6: S ) ← S ) ∪ { Si }
7: return S)
This is basically the same general notion as was used in Uniform Crossover or Bit-flip Mutation.
But you might not like this distribution of subsets. An alternative would be to pick a random
number under some distribution of your choosing and select a subset of that size:
Algorithm 50 Select a Subset (Second Technique)

1: S ← original set
2: n ← number of elements in the subset
3: subset S) ← {}
5: S) ← S) ∪ {random element from S chosen without replacement}
6: return S)
Note that unlike most situations here, we’re picking without replacement — that is, an element
can’t be picked more than once.
So back to crossover. One naive approach might be to pick some subset of nodes and subset
of edges in each graph, and exchange subsets. But what if graph A hands graph B an edge i → j
but B doesn’t have i or j among its nodes? Back to the drawing board. An alternative might be to
swap nodes, then swap edges with the constraint that an edge can only be swapped to the other
graph if the other graph received the relevant nodes as well. The difficulty here is, of course, that
the swapped-in subgraph will be disjoint with the existing nodes in that individual’s graph. And
you might miss some important edges that connected the nodes in the original graph.
68
A third choice is to pick whole subgraphs and swap them. To pick a subgraph:
Algorithm 51 Select a Subgraph

1: N ← nodes in the original graph
2: E ← edges in the original graph
3: N ) ⊆ N ← nodes in the subgraph (chosen with a subset selection operation)

4: subset E) ← {}
5: for each edge Ei ∈ E do
6: j, k ← nodes connected by Ei
7: if j ∈ N ) and k ∈ N ) then
8: E) ← E) ∪ { Ei }
9: return N ) , E)
Again, the problem is that the swapped-in subgraph is disjoint with the graph that’s already
there. At this point you may need to merge some nodes in the original graph with those in the
newly-swapped in subgraph. As nodes get merged together, certain edges need to be renamed
since they’re pointing to things that don’t exist any more. It’s still possible that the two graphs will
be disjoint but unlikely. We can force at least one node to merge, thus guaranteeing that the graphs
won’t be disjoint. The algorithm would then look something like this:
Algorithm 52 Randomly Merge One Graph Into Another

1: N ← nodes in the first graph, shuffled randomly " To shuffle an array randomly, see Algorithm 26
2: N ) ← nodes in the second graph
3: E ← edges in the first graph
4: E) ← edges in the second graph
5: p ← probability of merging a given node from N into a node from N )
6: for l from 1 to || N || do
7: if l = 1 or p ≥ random number chosen uniformly from 0.0 to 1.0 inclusive then
8: n) ← random node chosen uniformly from N ) " We’ll merge Nl with n)
9: for i from 1 to || E|| do
10: j, k ← nodes connected by Ei
11: if j = Nl then
12: Change j to n) in Ei
13: if k = Nl then
14: Change k to n) in Ei
15: else " No merge, just add Nl into the new graph directly
16: N ) ← N ) ∪ { Nl }
17: E) ← E) ∪ E
18: return N ) , E)
69
A final strategy, used in the N EAT algorithm, merges all the edges of two parents into one child.
But if edges have the same birthday (that is, originally they were the same edge), N EAT throws one
of them out. Thus subgraphs don’t just get arbitrarily merged during crossover: they’re merged
back in the way they used to be originally. The idea is to retain subgraph structures and reduce the
randomness of crossover.
We’ve not even gotten to how to make sure that your particular graph constraint needs (no
self-loops, no multiple edges, etc.) are kept consistent over crossover or mutation. What a mess.
4.3 Trees and Genetic Programming

Genetic Programming (GP) is a research community more than a technique per se. The community
focuses on how to use stochastic methods to search for and optimize small computer programs or
other computational devices. Note that to optimize a computer program, we must allow for the
notion of suboptimal programs rather than programs which are simply right or wrong.48 GP is thus
generally interested in the space where there are lots of possible programs (usually small ones)
but it’s not clear which ones outperform the others and to what degree. For example, finding
team soccer robot behaviors, or fitting arbitrary mathematical equations to data sets, or finding
finite-state automata which match a given language.
Because computer programs are variable in size, the representations used by this community
are also variable in size, mostly lists and trees. In GP, such lists and trees are typically formed
from basic functions or CPU operations (like + or if or kick-towards-goal). Some of these operations
cannot be performed in the context of other operations. For example, 4 + kick-towards-goal() makes
no sense unless kick-towards-goal returns a number. In a similar vein, certain nodes may be restricted
to having a certain number of children: for example, if a node is matrix-multiply, it might be
expecting exactly two children, representing the matrices to multiply together. For this reason, GP’s
initialization and Tweaking operators are particularly concerned with maintaining closure, that is,
producing valid individuals from previous ones.
One of the nifty things about optimizing computer programs is how you assess their fitness:
run them and see how they do! This means that the data used to store the genotypes of the
individuals might be made to conveniently correspond to the code of the phenotypes when run. It’s
not surprising that the early implementations of GP all employed a language in which code and
data were closely related: Lisp.
The most common form of GP employs trees as its representation, and was first proposed by
Nichael Cramer,49 but much of the work discussed here was invented by John Koza, to whom a lot
of credit is due.50
48 John Koza proposed exactly this notion in his book Genetic Programming: “...you probably assumed I was talking
about writing a correct computer program to solve this problem.... In fact, this book, focuses almost entirely on incorrect
programs. In particular, I want to develop the notion that there are gradations in performance among computer programs.
Some incorrect programs are very poor; some are better than others; some are approximately correct; occasionally, one
may be 100% correct.” (p. 130 of John R. Koza, 1992, Genetic Programming: On the Programming of Computers by Means of
Natural Selection, MIT Press.)
49 In a single paper Cramer proposed both tree-based GP and a list-based GP similar to that discussed in Section 4.4. He
called the list-based version the JB Language, and the tree-based version the TB Language. Nichael Lynn Cramer, 1985,
A representation for the adaptive generation of simple sequential programs, in John J. Grefenstette, editor, Proceedings of
an International Conference on Genetic Algorithms and the Applications, pages 183–187.
50 Except as noted, the material in Section 4.3 is all due to John Koza. For the primary work, see Footnote 48, p. 70.
70
Consider the tree at right,
√ containing the mathematical expres- sin
sion sin(cos( x − sin x ) + x x ). This is the parse tree of a simple
program which performs this expression. In a parse tree, a node +
is a function or if statement etc., and the children of a node are
the arguments to that function. If we used only functions and no cos ∗
operators (for example, using a function subtract(x, y) instead of
x − y ), we might write this in pseudo-C-ish syntax such as: − x sqrt
sin( x sin x
add(
x
cos(subtract(x, sin(x))),
multiply(x, sqrt(x)) Figure 18 A Symbolic Regression
)); tree.
The Lisp family of languages is particularly adept at this. In Lisp, the function names are tucked
inside the parentheses, and commas are removed, so the function foo(bar, baz(quux)) appears as
(foo bar (baz quux)). In Lisp objects of the form ( ... ) are actually singly-linked lists, so Lisp can
manipulate code as if it were data. Perfect for tree-based GP. In Lisp, Figure 18 is:
(sin
(+
(cos (− x (sin x)))
(∗ x (sqrt x))))
How might we evaluate the fitness of the individual in Figure 18? Perhaps this expression is
meant to fit some data as closely as possible. Let’s say the data is twenty pairs of the form $ xi , f ( xi )%.
We could test this tree against a given pair i by setting the return value of the x operator to be xi ,
then executing the tree, getting the value vi it evaluates to, and computing the some squared error
that is, δi = (vi − f ( xi ))2 . The fitness of an individual might be the square root of the
from f ( xi ),√
total error, δ1 + δ2 + ... + δn . The family of GP problems like this, where the objective is to fit an
arbitrarily complex curve to a set of data, is called symbolic regression.
Programs don’t have to be equations: they can actually do
things rather than simply return values. An example is the tree if-food-ahead
shown in Figure 19, which represents a short program to move an
ant about a field strewn with food. The operator if-food-ahead takes forward do
two children, the one to evaluate if there is food straight ahead,
and the one to evaluate if there isn’t. The do operator takes two left if-food-ahead
children and evaluates the left one, then the right one. The left and
right operators turn the ant 90◦ to the left or right, forward moves do right
the ant forward one step, consuming any food directly in front.
Given a grid strewn with food, the objective is to find a program forward left
which, when executed (perhaps multiple times), eats as much food Figure 19 An Artificial Ant tree.
as possible. The fitness is simply the amount of food eaten. This
is actually a common test problem called the artificial ant.
71
The code for Figure 19 in a pseudo-Lisp and C would look something like:
if (foodAhead) forward();
(if-food-ahead else {
forward left();
(do if (foodAhead) {
Pseudo-Lisp: left Pseudo-C: forward();
(if-food-ahead left();
(do forward left) }
right))) else right();
}
Tree-style GP can be used with any optimization algorithm of course. But for no particular
reason it has its own traditional algorithm, which was described earlier in Section 3.3.3.
GP builds new trees by repeatedly selecting from a function set (the collection of items which may
appear as nodes in a tree) and stringing them together. In the Artificial Ant example, the function
set might consist of if-food-ahead, do, forward, left, and right. In the Symbolic Regression example,
the function set might consist of +, −, ∗, sin, cos, sqrt, x, and various other mathematical operators.
Note that the functions in the function set each have an arity, meaning, a pre-defined number of
children. sin takes one child. do and + take two children. x and forward take no children. Nodes
with a zero arity (taking no children) are considered to be leaf nodes in the function set, and nodes
with an arity ≥ 1 are nonleaf nodes. Algorithms which string nodes together generally need to
respect these conventions in order to build a valid tree.
One common algorithm is the Grow algorithm, which builds random trees depth-first up to a
certain depth:
Algorithm 53 The Grow Algorithm

1: max ← maximum legal depth
2: FunctionSet ← function set
3: return DoGrow(1, max, FunctionSet)
4: procedure DoGrow(depth, max, FunctionSet)

5: if depth ≥ max then
6: return Copy(a randomly-chosen leaf node from FunctionSet)
7: else
8: n ← Copy(a randomly-chosen node from the FunctionSet)
9: l ← number of child nodes expected for n
11: Child i of n ← DoGrow(depth + 1, max, FunctionSet)
12: return n
72
The Full algorithm is a slight modification of the Grow algorithm which forces full trees up to the
maximum depth. It only differs in a single line:
Algorithm 54 The Full Algorithm

1: max ← maximum legal depth
3: return DoFull(1, max, FunctionSet)
4: procedure DoFull(depth, max, FunctionSet)

5: if depth ≥ max then
7: else
8: n ← Copy(a randomly-chosen non-leaf node from FunctionSet) " The only difference!
9: l ← number of child nodes expected for n
11: Child i of n ← DoFull(depth + 1, max, FunctionSet)
12: return n
GP originally built each new tree by picking either of these algorithms half the time, with a max
depth selected randomly from 2 to 6. This procedure was called Ramped Half-and-Half.
Algorithm 55 The Ramped Half-and-Half Algorithm

1: minMax ← minimum allowed maximum depth
2: maxMax ← maximum allowed maximum depth " ... if that name makes any sense at all...
4: d ← random integer chosen uniformly from minMax to maxMax inclusive

5: if 0.5 < a random real value chosen uniformly from 0.0 to 1.0 then
6: return DoGrow(1, d, FunctionSet)
7: else
8: return DoFull(1, d, FunctionSet)
The problem with these algorithms is that they provide no control over the size of the trees: and
indeed tend to produce a fairly odd distribution of trees. There are quite a number of algorithms
with better control.51 Here’s a one of my own design, PTC2,52 which produces a tree of a desired
size, or up to the size plus the maximum number of children to any given nonleaf node. It’s easy to
describe. We randomly extend the horizon of a tree with nonleaf nodes until the number of nonleaf
nodes, plus the remaining spots, is greater than or equal to the desired size. We then populate the
remaining slots with leaf nodes:
51 Liviu Panait and I did a survey of the topic in Sean Luke and Liviu Panait, 2001, A survey and comparison of tree
generation algorithms, in Lee Spector, et al., editors, Proceedings of the Genetic and Evolutionary Computation Conference
(GECCO-2001), pages 81–88, Morgan Kaufmann, San Francisco, California, USA.
52 PTC2 was proposed in Sean Luke, 2000, Two fast tree-creation algorithms for genetic programming, IEEE Transactions
on Evolutionary Computation, 4(3), 274–283. It’s an obvious enough algorithm that it’s been no doubt used many times
prior in other computer science contexts.
73
Algorithm 56 The PTC2 Algorithm
1: s ← desired tree size
3: if s = 1 then
5: else
6: Q←{}
7: r ← Copy(a randomly-chosen non-leaf node from FunctionSet)
8: c←1
9: for each child argument slot b of r do
10: Q ← Q ∪ {b}
11: while c + || Q|| < s do
12: a ← an argument slot removed at random from Q −
13: m ← Copy(a randomly-chosen non-leaf node from FunctionSet)
14: c ← c+1 + cos
15: Fill slot a with m
16: for each child argument slot b of m do ERC sin ∗
17: Q ← Q ∪ {b}
x sqrt x
18: for each argument slot q ∈ Q do
19: m ← Copy(a randomly-chosen leaf node from FunctionSet) ERC
20: Fill slot q with m
21: return r Figure 20 A tree with ERC
placeholders inserted. See
Figure 21.
Ephemeral Random Constants It’s often useful to include in the
function set a potentially infinite number of constants (like 0.2462 or −
$0.9, −2.34, 3.14% or 2924056792 or “s%&e : m”) which get sprinkled into
your trees. For example, in the Symbolic Regression problem, it might be + cos
nice to include in the equations constants such as -2.3129. How can we
do this? Well, function sets don’t have to be fixed in size if you’re careful. 0.231 sin ∗
Instead you might include in the function set a special node (often a leaf
node) called an ephemeral random constant (or ERC). Whenever an ERC x sqrt x
is selected from the function set and inserted into the tree, it automatically
transforms itself into a randomly-generated constant of your choosing. -0.194
From then on, that particular constant never changes its value again (un- Figure 21 The tree in Fig-
less mutated by a special mutation operator). Figure 20 shows ERCs ure 20 with ERC placehold-
inserted into the tree, and Figure 21 shows their conversion to constants. ers replaced with perma-
nent constants.
4.3.2 Recombination
GP usually does recombination using subtree crossover. The idea is straightforward: in each
individual, select a random subtree (which can possibly be the root). Then swap those two subtrees.
It’s common, but hardly necessary, to select random subtrees by picking leaf nodes 10% of the time
and non-leaf nodes 90% of the time. Algorithm 57 shows how select a subtree of a given type.
74
Algorithm 57 Subtree Selection
1: r ← root node of tree
2: f (node) ← a function which returns true if the node is of the desired type
3: global c ← 0
4: CountNodes(r, f )
5: if c = 0 then " Uh oh, no nodes were of the desired type!
6: return ! " “null” or “failure” or something
7: else
8: a ← random integer from 1 to c inclusive
9: c←0
10: return PickNode(r, a, f )
11: procedure CountNodes(r, f ) " This is just depth-first search

12: if f (r ) is true then
13: c ← c+1
14: for each child i of r do
15: CountNodes(i, f )
16: procedure PickNode(r, a, f ) " More depth-first search!

17: if f (r ) is true then
18: c ← c+1
19: if c ≥ a then
20: return r
21: for each child i of r do
22: v ← PickNode(i, a, f )
23: if v 4= ! then
24: return v
25: return ! " You shouldn’t be able to reach here
4.3.3 Mutation
GP doesn’t often do mutation,because the crossover operator is non-homologous53 and is highly
mutative. Even so, there are many possibilities for mutation. Here are just a few:
• Subtree mutation: pick a random subtree and replace it with a randomly-generated subtree
using the algorithms above. Commonly Grow is used with a max-depth of 5. Again, leaf
nodes are often picked 10% of the time and non-leaf nodes 90% of the time.
• Replace a random non-leaf node with one of its subtrees.
• Pick a random non-leaf node and swap its subtrees.
• If nodes in the trees are ephemeral random constants, mutate them with some noise.
53 Recall that with homologous crossover, an individual crossing over with itself will just make copies of itself.
75
• Select two subtrees in the individual such that neither is contained within the other, and swap
them with one another.
Again, we can use Algorithm 57 to select subtrees for use in these techniques. Algorithm 57 is
called subtree selection but it could have just as well been called node selection: we’re just picking a
node. First we count all the nodes of a desired type in the tree: perhaps we want to just select a leaf
node for example. Then we pick a random number a less than the number of nodes counted. Then
we go back into the tree and do a depth-first traversal, counting off each node of the desired type,
until we reach a. That’s our node.
4.3.4 Forests and Automatically Defined Functions

Genetic Programming isn’t constrained to a single tree: it’s perfectly reasonable to have a genotype
in the form of a vector of trees (commonly known as a forest). For example, I once developed
simple soccer robot team programs where an individual was an entire robot team. Each robot
program was two trees: a tree called when the robot was far from the ball (it returned a vector
indicating where to run), and another tree called when the robot was close enough to a ball to kick
it (it would return a vector indicating the direction to kick). The individual consisted of some n of
these tree pairs, perhaps one per robot, or one per robot class (goalies, forwards, etc.), or one for
every robot to use (a homogeneous team). So a soccer individual might have from 2 to 22 trees!
Trees can also be used to define automatically defined functions (ADFs)54 which can be called
by a primary tree. The heuristic here is one of modularity. Modularity lets us search very large
spaces if we know that good solutions in them are likely to be repetitive: instead of requiring the
individual to contain all of the repetitions perfectly (having all its ducks in order)-̇– a very unlikely
result — we can make it easier on the individual by breaking the individuals into modules with an
overarching section of the genotype define how those modules are arranged.
In the case of ADFs, if we notice that ideal solutions are likely to be large trees with often-
repeated subtrees within them, we’d prefer that the individual consist of one or two subfunctions
which are then called repeatedly from a main tree. We do that by adding to the individual a second
tree (the ADF) and including special nodes in the original parent tree’s function set55 which are just
function calls to that second tree. We can add further ADFs if needed.
For purposes of illustration, let’s say that a
good GP solution to our problem will likely need Main Tree ADF1 Tree
to develop a certain subfunction of two arguments. foo foo
We don’t know what it will look like but we be- foo ARG2
lieve this to be the case. We could apply this heuris- bar ADF1
tic belief by using a GP individual representation ARG1 bar
ADF1 quux quux
consisting of two trees: the main tree and a two-
argument ADF tree called, say, ADF1. baz quux ARG2
We add a new non-leaf node to its function
set: ADF1(child1, child2 ). The ADF1 tree can have quux
whatever function set we think is appropriate to Figure 22 An ADF example.
54 Automatically Defined Functions are also due to John Koza, but are found in his second book, John R. Koza, 1994,
Genetic Programming II: Automatic Discovery of Reusable Programs, MIT Press.
55 Every tree has its own, possibly unique, function set.
76
let GP build this subelement. But it will need to have two additional leaf-node functions added to
the main tree’s function set as well. Let’s call them ARG1 and ARG2.
Figure 22 shows an example individual. Here’s how it works. We first evaluate the main tree.
When it’s time to call an ADF1 node, we first call its two children and store away their results
(call them result1 and result2). We then call the ADF1 tree. When its ARG1 function is called, it
automatically returns result1. Likewise ARG2 automatically returns result2. When the ADF1 tree
is finished, we store away its return value (let’s call it final). We then return to the Main tree: the
ADF1 node returns the value final, and we continue execution where we left off in the Main tree.
Note that you could have more than one ADF tree. And you can have ADF trees which call
other ADF trees! There’s no reason you can’t have nested function calls, right? In theory you
could have recursive calls, that is, ADF trees which call each other. But your individuals won’t be
smart enough to build a base case automatically, so to keep the system from going into an infinite
recursive loop, you’ll need to have some maximum call depth built in.
One last variation: automatically defined macros (ADMs), due to Lee Spector.56 Here, when
the ADF1 node is called, we jump immediately to the ADF1 tree without bothering to call the
children to the ADF1 node first. Instead, whenever ARG1 is called, we jump back to the main tree
for a second, call the first child, get its result, come back to the ADF1 tree, and have ARG1 return
that value. This happens each time ARG1 is called. Likewise for ARG2. The idea is that this gives us
a limited ability to selectively, or repeatedly, call children, in a manner similar to if-then constructs,
while-loops, etc. (Lisp implements these as macros, hence the name).
4.3.5 Strongly-Typed Genetic Programming

Strongly-Typed Genetic Programming is a variant of Genetic Programming initially by David
Montana.57 Recall that in the examples shown earlier, each node returns the same kind of thing
(for example, in symbolic regression, all nodes return floating-point values). But in more complex
programs, this isn’t really an option. For example, what if we wanted to add to symbolic regression
a special operator, If, which takes three arguments: a boolean test, the then-value to return if the
test is true, and the else-value to return if the test is false. If returns floating-point values like the
other nodes, but it requires a node which returns a boolean value. This means we’ll need to add
some nodes which return only boolean values; both leaf nodes and perhaps some non-leaf node
operators like And or Not.
The problem here is that in order to maintain closure, we can no longer just build trees, cross
them over, or mutate them, without paying attention to the which nodes are permitted to be
children of which other nodes and where. What happens, for example, if we try to multiply
sin(x) by “false”? Instead, we need to assign type constraints to nodes to specify which nodes are
permitted to hook up with which others and in what way.
There are a variety of approaches to typing. In the simplest approach, atomic typing, each type
is just a symbol or integer. The return value of each node, the expected child types for each node,
and the expected return type for the tree as a whole, each get assigned one of these types. A node
may attach as the child of another, or act as the root node of the tree, only if the types match. In set
typing, the types aren’t simple symbols but are sets of symbols. Two types would match if their
56 Lee Spector, 1996, Simultaneous evolution of programs and their control structures, in Peter J. Angeline and K. E.
Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter 7, pages 137–154, MIT Press.
57 David Montana, 1995, Strongly typed genetic programming, Evolutionary Computation, 3(2), 199–230.
77
intersections are nonempty. Set typing can be used to provide sufficient typing information for a
lot of things, including the class hierarchies found in object-oriented programming.
But even this may not be enough. Atomic and set typing presume a finite number of symbols.
How would we handle the situation where nodes operate over matrices? For example, consider a
matrix-multiply node which takes two children (providing matrices) and multiplies them, returning
a new matrix. The dimensions of the returned matrix are functions of the two children matrices.
What if we change one of the children to a subtree which returns a new, differently-sized matrix? It’s
possible to do this if we can reconcile it by changing the return type of the parent. This may trigger
a cascade of changes to return types, or to the types of children, as the tree readjusts itself. Such
typing is commonly known as polymorphic typing and relies on type resolution algorithms similar
those found in polymorphic typing programming languages like Haskell or ML. It’s complex.
4.3.6 Cellular Encoding

Trees can also be used as short programs
a. parent b. E
to instruct an interpreter how to create a
second data structure (usually a graph). This
Double
second data structure is then used as the F
phenotype. This technique is commonly
left child right child c. E
known as Cellular Encoding (by Frédéric
58
Gruau). The general idea is to take a seed
Figure 23 The double Edge Encoding operator.
(perhaps a graph consisting of a single node
or a single edge) and hand it to the root of
the tree. The root operator modifies and expands the graph, then hands certain expanded elements
off to its children. They then expand the graph further, handing expanded pieces to their children,
and so on, until the tree is exhausted. The fully expanded graph is then used as the phenotype.
Gruau’s original formulation, which was used mostly for neural networks, operated on graph
nodes, which requires a fairly complicated mechanism. An alternative would be to operate on graph
edges, which doesn’t allow all possible graphs, but is fairly useful for sparse or planar graphs such
as are often found in electrical circuits or finite-state automata. Early on, Lee Spector and I dubbed
this Edge Encoding.59 Edge Encoding is easier to describe, so that’s what I’ll show off here.
Edge and Cellular Encoding tree nodes work differently from, say, the ones used for Symbolic
Regression: they take things from parents, operate on them, and then hand them to their children.
As an example, Figure 23 shows an Edge Encoding operator called double. It takes an edge handed
to it by its parent (Edge E in Figure 23b), and creates a duplicate edge connecting the same two
nodes (Edge F in Figure 23c). It then hands one edge each to its two children.
58 Frédéric Gruau, 1992, Genetic synthesis of boolean neural networks with a cell rewriting developmental process, in
J. D. Schaffer and D. Whitley, editors, Proceedings of the Workshop on Combinations of Genetic Algorithms and Neural Networks
(COGANN92), pages 55–74, IEEE Computer Society Press.
59 Lee Spector and I wrote an early paper on which named it Edge Encoding: Sean Luke and Lee Spector, 1996, Evolving
graphs and networks with edge encoding: Preliminary report, in John R. Koza, editor, Late Breaking Papers at the Genetic
Programming 1996 Conference, pages 117–124, Stanford Bookstore. But I doubt we’re the inventors: when the paper
came out, John Koza, Forrest Bennett, David Andre, and Martin Keane were already using a related representation to
evolve computer circuits. See John R. Koza, Forrest H Bennett III, David Andre, and Martin A. Keane, 1996, Automated
WYWIWYG design of both the topology and component values of electrical circuits using genetic programming, in
John R. Koza, et al., editors, Genetic Programming 1996: Proceedings of the First Annual Conference, pages 123–131, MIT Press.
78
1
a. c. e. !
0
1
b. d. ! f. ! 0 1
0 0
Figure 25 Expansion of a finite-state automaton using the Edge Encoding in Figure 24. (a) The initial edge. (b) After
applying double. (c) After applying reverse. (d) After applying loop, %, start, and 0. The white circle is a starting state. (e)
After applying bud and 1. (f) After applying split, 0, accept, and 1. The black circle is an accepting state.
Figure 24 shows an edge encoding tree which will construct double

a finite-state automaton. Besides double, the main operators are:
reverse, which reverses an edge; loop, which creates a self-loop reverse bud
edge on the head node of loop’s edge; bud, which creates a new
node and then a new edge from the head node of bud’s edge loop split 1
out to the new node; split, which splits its edge into an edge
! start 0 accept
from split’s tail node out to a new node, and then another edge
back to split’s head node. Other finite-state automaton-specific 0 1
operators (%, 1, 0) label their edge or (start, accept) label the
head node of their edge. Figure 24 An Edge Encoding.
Confused at this point? I would be! Perhaps this will help.
Figure 25 shows the expansion of Figure 24, starting with a single edge, and eventually growing
into a full finite-state automaton which interprets the regular language (1|0)∗ 01.
Cellular and Edge encoding are examples of an indirect or developmental encoding: a repre-
sentation which contains a set of rules to develop a secondary data structure which is then used
as the phenotype. Indirect encodings are a popular research topic for two reasons. First there’s
the biological attraction: DNA is an indirect encoding, as it creates RNA and protein which then
go on to do the heavy lifting in living organisms. Second, there’s the notion of compactness and
modularity discussed earlier: many indirect encoding rules make repeated calls to sub-rules of
some form. In Cellular and Edge encoding there’s no modularity, but you can add it trivially by
including some Automatically Defined Functions. Likewise, unless you use an ADF, there’s little
compactness: Edge Encoding trees will have at least as many nodes as the graph has edges!
4.3.7 Stack Languages

An alternative to Lisp are stack languages in which code takes the form of a stream of instructions,
usually in postfix notation. Real-world stack languages include F ORTH and PostScript. These
languages assume the presence of a stack onto which temporary variables, and in some cases
chunks of code, can be pushed and popped. Rather than say 5 × (3 + 4), a stack language might
say 5 3 4 + ×. This pushes 5, 3, and 4 on the stack; then pops the last two numbers (4 and 3) off the
stack, adds them, and pushes the result (7) on the stack; then pops off of the stack the remaining
two numbers (7 and 5), multiplies them, and pushes the result (35) back on.
79
Stack languages often create subroutines by pushing chunks of code (
onto the stack, then executing them from the stack multiple times. For
example, we might generalize the procedure above — a × (b + c) — into a ( ( (
subroutine by wrapping its operators in parentheses and subjecting them
to a special code-pushing operator like this: push (+×). Given another a b (
special operator do, which pops a subroutine off the stack, executes it n
times, and pushes it back on the stack, we can do stuff like 5 7 9 2 4 3 6 5 9 c
push (+×) 4 do, which computes 5 × (7 + 9) × (2 + 4) × (3 + 6) × (5 + 9).
Figure 26 The expression
Stack languages have long been used in genetic programming. Among ((a b) ( ) ((c))) as rooted
the most well-known is Lee Spector’s GP stack language, Push.60 Push parentheses.
maintains multiple stacks, one for each data type, allowing code to oper-
ate over different kinds of data cleanly. Push also includes special stacks "
for storing, modifying, and executing code. This allows Push programs
to modify their own code as they are executing it. This makes possible, for " "
example, the automatic creation of self-adaptive breeding operators.
a " ! "
The use of stack languages in optimization presents some represen-
tational decisions. If the language simply forms a stream of symbols with b ! " !
no constraints, just use a list representation (see the next Section, 4.4). But
most stack languages at least require that the parentheses used to delimit " !
code must be paired. There are many ways to guarantee this constraint.
In some stack languages a left parenthesis must always be followed by a c !
non-parenthesis. This is easy to do: it’s exactly like the earlier Lisp expres- Figure 27 The expression
sions (see Figures 18 and 19). If instead your language allows parentheses ((a b) ( ) ((c))) in cons cells.
immediately after left parentheses, as in ((a b) ( ) ((c))), you could just
use the left parenthesis as the root node of a subtree and the elements inside the parentheses as the
children of that node, as shown in Figure 26. Both approaches will require that tree nodes have
arbitrary arity. Or, as is the case for Push, you could use the traditional internal format of Lisp:
nested linked lists. Each parenthesized expression (like (a b)) forms one linked list, and elements
in the expression can be other linked lists. Nodes in each linked list node are called cons cells,
represented in Figure 27 as ". The left child of a cons cell holds a list element, and the right child
points to the next cons cell in the list, or to !, indicating the end of the list.
4.4 Lists, Machine-Language Genetic Programming, and Grammatical Evolution

Parse trees aren’t the only way to represent programs. Some GP practitioners represent programs
as arbitrary-length lists (or strings) of machine language instructions. Individuals are evaluated by
converting the lists into functions and executing them. The is commonly known as Linear Genetic
Programming, and the most famous practitioners of this approach are Wolfgang Banzhaf, Peter
Nordin, Robert Keller, and Frank Francone, who sell a superfast GP system called Discipulus based
on this notion, and who also wrote a well-regarded book on Genetic Programming which was half
tree-based and half linear GP.61
60 The basics: Lee Spector and Alan Robinson, 2002, Genetic programming and autoconstructive evolution with the
push programming language, Genetic Programming and Evolvable Machines, 3(1), 7–40. Then for the latest version of the
language, check out: Lee Spector, Jon Klein, and Martin Keijzer, 2005, The Push3 execution stack and the evolution of
control, in Proceedings of the Genetic and Evolutionary Conference (GECCO 2005), pages 1689–1696, Springer.
61 Wolfgang Banzhaf, Peter Nordin, Robert E. Keller, and Frank D. Francone, 1998, Genetic Programming: An Introduction,
Morgan Kaufmann.
80
Grammar An Arbitrary Individual
tree → n + n | n−n
n → n∗m | sin m false false true true false true false true true false false
m → 1 | 2
Figure 28 A Grammatical Evolution grammar, and an individual with a list representation.
Interpreting... [start] false false true true false true false
Expansion tree + + + + + + +
n n ∗ n ∗ n ∗ n ∗ n ∗ sin ∗ sin
n m sin m sin m sin 1 sin 1 m sin 1 1
m 2 2 2 2
Figure 29 Expansion of the individual shown in Figure 28.
Executing arbitrary machine code strings can be dangerous if closure isn’t maintained. But how
to maintain closure in such a situation? Certainly your individual wouldn’t be just a bit-string,
because that would allow all sorts of machine language instructions, even undesirable ones or
nonsense ones.62 Clearly it’d have to be a list of instructions chosen from a carefully-selected set.
If the instruction set is finite in length, we could just assign a unique integer to each instruction
and represent a genotype as a list of integers. Usually schemes employ a finite set of registers as
well: this allows the machine code lists to operate essentially like directed acyclic graphs (DAGs),
with early instructions affecting instructions much further down in the list due to their shared
register. Additionally we might find it desirable to include some special instructions that operate
on constants (Add 2, etc.).
Stack languages bear a strong resemblance to machine code, so it shouldn’t be too terribly
surprising that, as mentioned in Section 4.3.7, some stack languages are straightforwardly applied
to list representations, particularly if the language has no particular syntactic constraints.
Lists can be used to generate trees as well: consider a final GP scheme, Grammatical Evolution
(GE), invented by Conor Ryan, J. J. Collins, and Michael O’Neill.63 Grammatical Evolution’s
representation is a list of integers or boolean values. It then uses this list as the decision points in a
pre-defined tree grammar to build a GP Tree. The tree is then evaluated in GP style to assess fitness.
This somewhat complex approach is yet another example of an indirect encoding, and though it
doesn’t have the modularity common in many indirect encodings, it does have a method to its
madness: it can straightforwardly define any tree for any desired language.
As an example, consider the ridiculous grammar and an individual represented as a list, shown
in Figure 28. To interpret this, we start with tree, and use the first element in the list to decide how
to expand that (we’ll assume that false expands to the first item, and true expands to the second
item). Once we expand, we expand the remaining undefined variables in a depth-first fashion.
Figure 29 shows the expansion of the Individual in Figure 28.
62 You
probably don’t want to call the infamous HCF (“Halt and Catch Fire”) instruction. Look for it on Wikipedia.
63 Conor Ryan, J. J. Collins, and Michael O’Neill, 1998, Grammatical evolution: Evolving programs for an arbitrary
language, in EuroGP 1998, pages 83–96.
81
Now we have a tree we can evaluate! Notice that we wound up not using the last 4 bits in the
individual (true true false false). What if the list is too short and we don’t have enough decision
points? Typically one just wraps around to the beginning of the list again. It’s not a great solution
but it’s workable.64 GE is clever in that it allows us to construct any valid tree for a given grammar,
which is a lot more flexible than standard Tree-based GP: indeed it negates the need to even bother
with strong typing. The downside is that this representation is naturally un-smooth in certain
places: tiny changes early in the list result in gigantic changes in the tree. This can be a problem.
How new lists are generated largely depends on the domain-specific needs of the method involved.
But generally speaking there are two issues: specifying the length of the list, and populating it. One
simple way to do the former is to sample a length from the geometric distribution (Algorithm 46,
perhaps with the minimum list size being 1). Beware again that the distribution will have a very
high number of small lists: you may wish to use a flatter distribution.
To populate the list, just march through the list and set each of its values to something random
but appropriate. Remember that for some problems this isn’t sufficient, as there may be constraints
on which elements may appear after other elements, so you’ll need to be more clever there.
4.4.2 Mutation
Like initialization, mutation in lists has two parts: changing the size of the list, and changing the
contents of the list. Contents may be changed in exactly the same way that you do for fixed-length
vectors: using a bit-flip mutation or integer randomization, etc. Remember that you may not be able
to change some elements without changing others due to certain constraints among the elements.
Changing the length likewise depends on the problem: for example, some problems prefer to
only add to the end of a list. One simple approach is to sample from some distribution, then add
(or subtract, if it so happens) that amount to the list length. For example, we could do a random
walk starting at 0, flipping a coin until it comes up tails. The number we arrive at is what you add
to (or delete from, if it’s negative) the end of the list. This should look familiar:
Algorithm 58 Random Walk

1: b ← coin-flip probability " Make b bigger to make the random walks longer and more diffuse
2: m←0
4: repeat
5: n ← either a 1 or -1, chosen at random.
6: if m + n is an acceptable amount then
7: m ← m+n
8: else
9: m ← m−n
10: until b < random number chosen uniformly from 0.0 to 1.0 inclusive
11: return m
64 I don’t like that approach: instead, I’d bypass evaluation and just assign the individual the worst possible fitness.
82
Don’t confuse this with Algorithm 42 (Random Walk Mutation), which uses a similar random
walk to determine the noise with which to mutate. Beware that because lists can’t be any smaller
than 1, but can be arbitrarily large, a random walk like this may cause the individual lists to become
fairly large: you may need to add some countering force to keep your population from growing
simply due to your mutation operator (see the bloat discussion below for other reasons for growth).
Warning In some list-representation problems, such as Grammatical Evolution, the early elements
in the list are far more important than the later elements. In GE this is because the early elements
determine the early choices in the tree grammar, and changing them radically changes the tree;
whereas the later elements only change small subtrees or individual elements (or if the list is too
long, they don’t change anything at all!) This has a huge effect on the smoothness of the landscape,
and you want to make sure your mutation procedure reflects this. For example, you might only
occasionally change the elements at the beginning of the list, and much more often change the
elements near the end of the list. Linear GP may or may not have this property depending on
the nature of your problem, and in fact it can actually can have the opposite situation if the final
machine code elements in the list get to make the last and most important changes.
4.4.3 Recombination
Like mutation, crossover also may depend on constraints, d
but ignoring that, there are various ways you could do
0 1 0 0
crossover among variable-length lists. Two easy ones are
one-point and two-point list crossover, variations on the Swap
standard one- and two-point vector crossovers. In one-

point list crossover, shown in Figure 30, we pick a (possibly 0 0 1 0 1 1 0 0 0
different) point in each list, then cross over the segments
to the right of the points. The segments can be non-zero in c
length. The algorithm should look eerily familiar:
Figure 30 One-point List Crossover.
Algorithm 59 One-Point List Crossover

v ← first list $v1 , v2 , ...vl % to be crossed over
1: !
! ← second list $w1 , w2 , ...wk % to be crossed over
2: w

4: d ← random integer chosen uniformly from 1 to k inclusive
5: !x ← snip out vc through vl from !v
6: !y ← snip out wd through wk from w !
7: Insert !y into !v where !x was snipped out
8: Insert !x into w! where !y was snipped out
9: return !v and w !
Two-point list crossover, shown in Figure 31, is similar: we pick two points in each individual
and swap the mid-sections. Again, note that the points don’t have to be the same. Think carefully
about your list representation to determine if one- or two-point list crossover make sense. They
have quite different dynamics. Is your representation reliant to the particulars of what’s going on
in the middle, and sensitive to disruption there, for example?
83
Another Warning Just as mentioned for mutation, certain elements of the list may be more
important than others and more sensitive to being messed up via crossover. So in Grammatical
Evolution for example you might want to consider picking two-point crossover points near to the
end of the list more often than ones near the front. Or stick with one-point crossover.
The two-point list crossover algorithm should likewise feel familiar to you:
Algorithm 60 Two-Point List Crossover

v ← first list $v1 , v2 , ...vl % to be crossed over
1: !
! ← second list $w1 , w2 , ...wk % to be crossed over
2: w

4: d ← random integer chosen uniformly from 1 to l inclusive
5: e ← random integer chosen uniformly from 1 to k inclusive
6: f ← random integer chosen uniformly from 1 to k inclusive
e f
7: if c > d then
8: Swap c and d
0 0 1 0 0
9: if e > f then
10: Swap e and f Swap
11: !x ← snip out vc through vd from !v

0 0 1 0 1 1 0 0 0
12: !y ← snip out we through w f from w !
13: Insert !y into !v where !x was snipped out c d
14: Insert !x into w! where !y was snipped out
Figure 31 Two-point List Crossover.
15: return !v and w !
4.5 Rulesets
A set is, of course, a collection of objects, possibly empty, where all the objects are different. Sets can
be used for all sorts of stuff, but the big item seems to be sets of rules which either form a computer
program of sorts (perhaps to direct a robot about in a simulated environment) or which define an
indirect encoding which grows a graph structure from a simple initial seed.
Rules in rulesets usually take a form which looks like if→then. The if part is commonly called
the body of the rule and the then part is commonly called the head of the rule. There are two
common kinds of rulesets, which I will call state-action and production rulesets. State-action rules
are designed to perform some action (the then) when some situation or event has occurred in the
world (the if ). For example, a robot’s sensors might trigger a rule which causes the robot to turn left.
Production rules are different in that some rules’ then actions trigger other rules’ if portions. For
example, if a rule a → b fires, it would then cause some other rule b → c to fire. Production rules are
mostly used to construct indirect encodings which grow graph structures etc. The interconnection
among the rules in production rulesets means that they bear more than a passing resemblance,
representation-wise, to directed graph structures.
The first question is: what data structure would you use to hold a set of objects? We could use a
variable-sized vector structure like a list. Or we could use a hash table which stores the elements as
keys and arbitrary things as values. In my experience, most people implement sets with lists.
The basic closure constraint in a set is its uniqueness property: often you have to make sure
that when you create sets, mutate them, or cross them over, the rules remain all different. Unless
84
you have a mutation or crossover operation which does this naturally, you may need to go back
into the set after the fact and remove duplicates. This is a trivial procedure:
Algorithm 61 Duplicate Removal

v ← collection of elements converted into a vector $v1 , v2 , ...vl %
1: !
2: h ← {} " Represent h with a hash table, it’s faster

3: l) ← l
4: for i from l down to 1 do
5: if vi ∈ h then " A duplicate!
6: Swap vi and vl )
7: l) ← l) − 1
8: else " Not a duplicate!
9: h ← h ∪ { vi }
10: !v ← blank vector $v1) , v2) , ...v)l ) %
)
11: for i from 1 to l ) do

12: vi) ← vi
13: return !v) converted back into a collection
Note that this modifies the order of the original list !v. You can represent h with a hash table
easily: to add an element to h, you just add it as the key to the hash table (the value can be anything:
for example, the element itself). To test to see if vl ∈ h, you just check to see if vl is a key in the hash
table already. Piece of cake.
4.5.1 State-Action Rules

An agent is an autonomous computational entity, that is, one which manipulates the world on its
own, in response to feedback it receives from the world. Agents include autonomous robots, game
agents, entities in simulations, etc. One common kind of program an agent might follow is a policy:
a collection of simple rules to tell the agent what to do in each possible situation it may find itself
in. These rules are often called state-action rules. Here are some state-action rules for an agent to
get around in a city: are you downtown? Then get on the train. Are you on the train? Then take the
train to the wharf. Did you miss your stop? Then get off the train and get on the return train. Etc.
State-action rules take on various guises, but a typical form is a ∧ b ∧ ... ∧ y → z, where the
a, b, ..., y are state descriptions and z is an action or class. A state description is some feature about the
current world that might or might not be true. An action is what we should do if that feature is
true. For example, a robot might have rules like:
Left Sonar Value > 3.2 ∧ Forward Sonar Value ≤ 5.0 −→ Turn Left to 50◦
We might test our ruleset by plopping a simulated robot down in an environment and using
these rules to guide it. Each time the robot gets sensor information, it gathers the rules whose
bodies are true given its current sensor values. The matching rules are collectively known as the
match set. Then the robot decides what to do based on what the heads of these rules suggest
(suggest as “turn left by 50◦ ”).
85
One way to think of the rule bodies is as describing regions in the state space of the robot, and
the heads as what to do in those regions. In the case of the rule above, the rule body has roped off a
region that’s less than 3.2 in one dimension and ≥ 5.0 in another dimension, and doesn’t cut out
any portions along any other dimensions.
There are two interesting issues involved here. First, what if no rules match the current
condition? This is commonly known as under-specification of the state space: there are holes in
the space which no rule covers. This is often handled by requiring a default rule which fires when
no other rule fires. More interestingly, what if more than one rule matches the current condition,
but those rules disagree in their heads in an incompatible way (one says “Turn Left” and one
says “Turn Right”, say)? This is known as over-specification of the state space. We’ll need employ
some kind of arbitration scheme to decide what to do. Most commonly, if we have lots of rules,
we might have a vote. Another way is to pick a rule at random. And yes: a state space can be
simultaneously under- and over-specified.
State-action rulesets often introduce a twist to the fitness assessment process. Specifically, as we
move the agent around, we may not only assess the fitness of the individual itself but also assess
the fitness of the individual rules inside the ruleset individual. At the very least this can be done by
breaking the rules into those which fired during the course of running the individual and those
which never fired (and thus aren’t responsible for the wonderful/terrible outcome that resulted).
We can then punish or reward only the rules which fired. Or if after turning Left the robot received
an electric shock, which might penalize the series of rules whose firings which led up to that shock,
but not penalize later rules. We might be more inclined to mutate or eliminate (by crossover) the
more-penalized rules.
Metaheuristics designed for optimizing policies using state-action rules, Michigan-Approach
Learning Classifier Systems and Pitt-Approach Rule Systems, are discussed in Section 10.
4.5.2 Production Rules

Production rules are similar to state-action rules except that the “actions” are used to trigger the
states of other rules. Production rules are sort of backwards looking: they tend to look like this:
a → b ∧ c ∧ ... ∧ z. This is because production rules are fired (typically) by a single event (triggered
by some other rule usually) and then this causes them to trigger multiple downstream rules in turn.
A lot of the uses for production rules is to enable modular indirect encodings which can describe
large complex solutions with lots of repetitions, but do so in a small, compact rule space which is
more easily searched. This of course assumes that good solutions will have lots of repetitions; this
in turn depends largely on the kind of problem you’re trying to solve.
Typically the symbols which appear in the heads (the right side) production rules are of two
forms: nonterminal symbols, which may also appear in the bodies of rules, and terminal symbols,
which often may not. Terminal symbols basically don’t expand any further. Note that for most
production systems, there’s a fixed number of rules, one per nonterminal.
An early example of applying evolutionary computation to production rules was developed
by Hiroaki Kitano to find certain optimal graph structures for recurrent neural networks and the
like.65 Imagine that you’re trying to create an 8-node directed, unlabeled graph structure. Our
ruleset might look like this (numbers are terminal symbols):
65 Thispaper was one of the seminal papers in indirect encodings. Hiroaki Kitano, 1990, Designing neural networks
using a genetic algorithm with a graph generation system, Complex Systems, 4, 461–476.
86
* + * + * + * + * + * +
b c 1 0 1 1 0 1 0 0 1 1
a→ b→ c→ d→ 0→ 1→
c d d c 1 0 0 0 0 0 1 1
This is an indirect encoding of the graph structure, believe it or not. We start with
* the
+ 1×1
, - , - b c
matrix a . We then apply the rule which matches a, expanding the a matrix into . From
c d
there we apply rules to each of the elements in that matrix, expanding them into their 2 × 2 elements,
 
1 1 0 0 1 1 1 1
1 1 0 0 1 1 1 1
   
1 0 1 1 0 1 1 1 1 1 0 0
 
 d c 1 0 0 0 1 0 1 1 0 0
resulting in the matrix  
1 1 0 1. From there, we expand to 1 1 1 1 0 0 1 1. At this

 
1 0 0 0 1 1 1 1 0 0 1 1
 
1 1 0 0 0 0 0 0
1 1 0 0 0 0 0 0
+ *
1 1
point we’re out of nonterminal symbols. (Since we made up “expansion rules” like 1 →
1 1
for our terminal symbols, we could have either expanded until we ran out of nonterminals, or
expanded some number of predefined times.) This is our adjacency matrix for the graph, where
a 1 at position $i, j% means “there’s an edge from i to j” and a 0 means “no edge”. I won’t bother
drawing this sucker for you!
A more recent example of indirect encoding with produc-
tion rules is in finding optimal Lindenmayer Systems (or L-
Systems). These are sets of production rules which produce
a string of symbols. That string is then interpreted as a small
computer program of sorts to produce some final object such
as a plant or tree, fractal or pattern, or machine of some sort.
L-Systems were made popular by Aristid Lindenmayer, a biol-
ogist who developed them to describe plant growth patterns.66 Figure 32 Plant patterns created by a Lin-
A simple example of an L-System is one which creates the denmayer System.
Koch Curve, a fractal pattern. The rule system consists of the single rule F → F + F − F − F + F.
It works like this: we start with a single F. Applying this rule, this expands to F + F − F − F + F.
Expanding each of these F’s using the rule, we get:
F+F−F−F+F+F+F−F−F+F−F+F−F−F+F−F+F−F−F+F+F+F−F−F+F
Expanding yet again, we get:
F + F − F − F + F + F + F − F − F + F − F + F − F − F + F − F + F − F − F + F + F + F − F − F + F+
F + F − F − F + F + F + F − F − F + F − F + F − F − F + F − F + F − F − F + F + F + F − F − F + F−
F + F − F − F + F + F + F − F − F + F − F + F − F − F + F − F + F − F − F + F + F + F − F − F + F−
F + F − F − F + F + F + F − F − F + F − F + F − F − F + F − F + F − F − F + F + F + F − F − F + F+
F+F−F−F+F+F+F−F−F+F−F+F−F−F+F−F+F−F−F+F+F+F−F−F+F
66 Przemyslaw Prusinkiewicz and Aristid Lindenmayer produced a beautiful book on L-Systems: Przemyslaw
Prusinkiewicz and Aristid Lindenmayer, 1990, The Algorithmic Beauty of Plants, Springer-Verlag. It’s out of print
but available online now, at http://algorithmicbotany.org/papers/#abop
87
The + and − are terminal symbols. What do you do with such a string? Well, if you interpreted
the F as “draw a line forward” and + and − as “turn left” and “turn right” respectively, you
would wind up with the Koch Curve shown in Figure 33. Further expansions create more complex
patterns.
These rules can get really complicated. Figure 34 shows
an actual L-System used by biologists to describe the
branching pattern of the red seaweed Bostrychia radicans.67
One interesting use of L-Systems with evolutionary
computation, by Greg Hornby, was in discovering useful
designs such as novel chairs or tables. Hornby also ap-
plied L-Systems together with Edge Encoding to discover
animal body forms and finite-state automata-like graph
structures.68 The L-System ruleset expanded into a string, Figure 33 A Quadratic Koch Curve.
which was then interpreted as a series of Edge Encoding
instructions (double, split, etc.) to produce the final graph.
Like direct-enoded graph structures, building rulesets is O → FGD
mostly a matter of determining how many elements you D → G[+++FGFGRG][-GF]GFGA
want, and then creating them. We begin by picking a de- A → FGFGFGFG[+++FGR][-GF]GFGB
B → FGFGFGFG[+++FGR][-GF]GFGC
sired ruleset size n, using some distribution (the Geometric C → FGFGFGFG[+++FGR][-GF]GFGK
Distribution, Algorithm 46, is probably fine). We then cre- R → FG[+FGFGU]GFGFGE
ate a ruleset out n of randomly-generated elements. E → [-FGFGX]GFGFGH
When doing production rules, there are some addi- H → [+FGFGW]GFGFGZFG
K → FGFGFG[+++FGR][-FGA]GFGL
tional constraints. Specifically, the various symbols which L → FGFGFG[+++FGR][-GF]GFGP
appear in the heads of the rules need to match symbols in P → FGFGFG[+++FGR][-GF]GFGQ
the bodies of the rules. Otherwise, how would you match Q → FGFGFGT
up an event triggered by a rule with the follow-on rule T → FGFGFG[+++FGR][+FGA]GFGA
which is fired as a result? Likewise, you probably won’t U → [+FGFGF]GFG
X → [-FGFGF]GFG
want two rules that have the same body, that is, two pro- W → [+FGFGF]GFG
duction rules of the form a → b, c and a → d, e, f . Which Z → [-FGFGF]GFG
one should fire? Arbitration doesn’t make much sense in
production rules, unlike state-action rules, unless perhaps Figure 34 Another L-System.
your production rules are probabilistic.

In some production rule systems, the number of rules is fixed to the size of the nonterminal set.
In other rule systems you might have a variable number of symbols. In the second case you will
need to make sure that all the symbols in rule heads have a corresponding rule with that symbol in
67 From Ligia Collado-Vides, Guillermo Gómez-Alcaraz, Gerardo Rivas-Lechuga, and Vinicio Gómez-Gutierrez, 1997,
Simulation of the clonal growth of Bostrychia radicans (Ceramiales-Rhodophyta) using Lindenmayer systems, Biosystems,
42(1), 19–27.
68 Greg gave what I consider the best paper presentation ever at GECCO. He did a regular presentation on using
L-systems to evolve walking creatures. But at the end of the presentation he dumped out of a canvas sack a mass of
tinkertoys and servos. He pressed a button, and it came to life and began to walk across the table. It was a tinkertoy
version of his best-fitness-of-run individual. For more information on Greg’s work, his thesis is a good pick: Gregory
Hornby, 2003, Generative Representations for Evolutionary Design Automation, Ph.D. thesis, Brandeis University.
88
the body. And rules with symbols in their bodies but appearing nowhere in any other rule’s heads
are essentially orphans (this can happen in the fixed-case as well). Additionally, you may or may
not allow recursion among your rules: can rule A trigger rule B, which then triggers rule A again?
For example, imagine if letters are our expansion variable symbols and numbers are our terminals.
Here’s a ruleset with some potential problems:
a → b, c There’s no c rule! What gets triggered from the c event?

b → d, 0
a → d, b Um, do we want duplicate rule bodies?
d → b Is recursion allowed in this ruleset?
e → 1 There’s nothing that will ever trigger this rule! It’s just junk!
During initialization you’ll need to handle some of these situations. You could generate rules at
random and then try to “fix” things. Or you could create some n nonterminal symbols and then
construct rules for each of them. Here’s an algorithm along those lines: it’s not particularly uniform,
but it does let you choose whether to allow recursive rules or not, and whether or not to permit
disconnected rules (that is, ones never triggered). It should get you the general idea: but if you
used this, you’d probably need to heavily modify it for your purposes.
Algorithm 62 Simple Production Ruleset Generation

1: !t ← pre-defined set of terminal symbols (that don’t expand)
2: p ← approximate probability of picking a terminal symbol
3: r ← flag: true if we want to allow recursion, else false
4: d ← flag: true if we want to allow disconnected rules, else false
5: n ← a random integer > 0 chosen from some distribution

6: !v ← vector of unique symbols $v1 , ..., vn % " The symbol in v1 will be our “start” symbol
−−→
7: rules ← empty vector of rules $rules1 , ..., rulesn %
8: for i from 1 to n do " Build rules
9: l ← a random integer ≥ 1 chosen from some distribution
10: !h ← empty vector of symbols $h1 , ...hl %
11: for j from 1 to l do
12: if (r = false and i = n) or p < random value chosen uniformly from 0.0 to 1.0 inclusive then
13: h j ← a randomly chosen terminal symbol from !t not yet appearing in !h
14: else if r = false then
15: h j ← a randomly chosen nonterminal from vi+1 , ..., vn not yet appearing in !h
16: else
17: h j ← a randomly chosen nonterminal symbol from !v not yet appearing in !h
18: rulesi ← rule of the form vi → h1 ∧ h2 ∧ ... ∧ hl
19: if d = false then " Fix disconnected rules
21: if vi does not appear in the head of any of the rules rules1 , ..., rulesi−1 then
22: l ← a random integer chosen uniformly from 1 to i − 1 inclusive
23: Change rulesl from the form vl → h1 ∧ ... to the form vl → vi ∧ h1 ∧ ...
−−→
24: return rules
89
4.5.4 Mutation
Mutation in sets is often similar to mutation in lists. That is, you usually have two tasks: changing
the size of the ruleset (if you’re allowed to), and mutating the rules in the set. The advice we have
to offer here is basically the same as for lists too. For example, one way to change the size is to
sample a small value from the geometric distribution, then either add or delete that number of
rules from the set (you might select victims at random). Likewise, you could mutate rules in the set
in the same manner as bit-flip mutation: mutate each rule with a certain independent probability.
Production rules, as usual, have additional constraints. If you mutate the head of a rule you’ll
need to make sure the resultant symbols match up. You may have created orphans, and will need
to decide if that’s acceptable. Likewise you may want to be very careful about mutating the body
(the primary symbol) of a production rule — you may create orphans or rulesets with more than
one rule with the same symbol in the body.
4.5.5 Recombination
If you have a fixed number of rules, one per production symbol (for example), then recombination
may be easy: just use uniform crossover to swap some subset of the rules. If your number of rules is
arbitrary, you may need to pick a subset of rules in each individual and swap them (see Algorithm
49 to select a subset).
In any case, if you’ve got constraints on your rules (such as in production rules), you need to be
careful about crossover: what happens if you orphan a rule? (Or do you care?) What happens if
you eliminated a rule for which some other rule had an event that triggered it? Who gets triggered
now? One of the biggest issues in crossing over arbitrary-length production rulesets is in merging
the symbols: you may have symbols in one ruleset which don’t match up with symbols in the
other ruleset. As a result, the rulesets are essentially disjoint. How do you merge them? You may
need to go through the rulesets and decide that certain rules in one ruleset will trigger rules in the
crossed-in ruleset in certain ways. This can be a real mess. And there isn’t any good guidance here:
like graphs, it’s fairly ad-hoc.
4.6 Bloat
Many of the representations presented here are variable-sized in nature. ...
One of the interesting problems with variable-sized representations is that,
... ×
over time, the individuals in your population may start to increase in size.
This is commonly known as bloat (or “code bloat”, or “code growth”, take + 0
your pick69 ). Bloat has been studied the most in the context of Genetic
Programming, where it’s a real problem. Bloated individuals are slow x x
to evaluate. They consume your memory. And worst of all, bloated
Figure 35 Inviable code ex-
individuals tend to be very far from the optimum, which is often not all ample.
that big. It’s a deceptive problem situation.
Early on GP researchers noticed that large, bloated trees contained a lot of subtrees which
didn’t do anything at all. These subtrees were dubbed introns, like their DNA counterparts. Figure
35 shows a particular kind of intron called inviable code. Here, the subtree (+ x x) is worthless
because no matter what it returns, it’s multiplied against zero. Early GP work assumed that these
69 Bill Langdon has been known to call it “survival of the fattest”.
90
introns were the problem. The idea went like this: most Tweaks are damaging to the fitness of
an individual. So in order to survive, you don’t really need to improve yourself per se: you just
need to be able to keep your head above water. One way to do this is to make it hard for Tweak to
damage you, even if that also means preventing it from improving you. If you have a lot of introns,
and particularly inviable code, then you increase the chance that the Tweaks will fall in those intron
regions, where it doesn’t matter what gets changed, and your fitness will stay the same. That was
the idea anyway. But it turned out to not be true: if you eliminated the ability for individuals to
Tweak in inviable code regions, bloat kept on going.70
So at this stage the reason for bloat is still up in the air. My own theory is that bloat is mostly
due to individuals wanting to have deeper and deeper Tweak points because deeper areas in the
tree tend to have less of an effect on the fitness of the child. Deeper points turn out to be well
correlated with bigger kids. There are other theories out there too.
At any rate, lacking a good firm understanding of just what the heck is going on, most of
the approaches to bloat control are somewhat ad-hoc. There are three common ways to keep
individuals small:
• Limit the sizes when the individuals are Tweaked.
• Edit the individuals, to remove introns and the like.
• Punish individuals for being very large.
Genetic Programming traditionally limited individuals by placing a bound on the maximum

allowable depth of an individual (to 17). But nowadays the trend is towards punishing individuals
for being large by hurting their chances of being selected. This is called parsimony pressure.71
The most straightforward way to do parsimony pressure is to include some portion of the indi-
vidual’s size as part of their fitness. For example, we might define the fitness f of an individual
as f = αr − (1 − α)s where r is the raw (or “actual”) fitness of the individual, and s is the size of
the individual. 0 ≤ α ≤ 1 is a constant to determine how much size counts and how much raw
fitness counts. This concept is popularly known as linear parsimony pressure. The big problem
with linear parsimony pressure is that you have to know how much of size is worth so much of
fitness. And as discussed before, we usually design fitness functions in an arbitrary way. Even so,
linear parsimony pressure is often quite good.
Another approach is to do a non-parametric parsimony pressure method which doesn’t con-
sider the specific fitness or fatness values of the individual and just picks based on comparing who’s
fitter and who’s fatter. There are lots of these. For example, lexicographic parsimony pressure
modifies Tournament selection to have individuals win if they’re fitter, but if there’s a tie, the
individual that wins is the one that’s smaller.72
70 Iwrote a paper which showed this: Sean Luke, 2000, Code growth is not caused by introns, in Darrell Whitley, editor,
Late Breaking Papers at the 2000 Genetic and Evolutionary Computation Conference, pages 228–235.
71 Liviu Panait and I did a shootout of all the parsimony pressure methods described here, and a host of other popular
methods, in Sean Luke and Liviu Panait, 2006, A comparison of bloat control methods for genetic programming,
72 Lexicographic parsimony pressure has been around since at least 1994, where it appeared casually in both Conor
Ryan, 1994, Pygmies and civil servants, in Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 11,
pages 243–263, MIT Press (and) Simon Lucas, 1994, Structuring chromosomes for context-free grammar evolution, in
Proceedings of the First IEEE Conference on Evolutionary Computation, pages 130–135, IEEE.
91
Algorithm 63 Lexicographic Tournament Selection
1: P ← population

6: if Fitness(Next) > Fitness(Best), or Fitness(Next) = Fitness(Best) and Next is smaller, then
7: Best ← Next
8: return Best
This works okay in environments where there are lots of ties. But often that’s not the case.
Another approach is to push for both smallness and fitness in your chosen individual. In double
tournament, we do a tournament selection based on fitness. But the individuals entering the
tournament aren’t from the general population. Rather they were chosen from other tournament
selection operations which were based on smallness.
Algorithm 64 Double Tournament Selection

1: P ← population
2: t1 ← tournament size for fitness, t1 ≥ 1
3: t2 ← tournament size for smallness, t2 ≥ 1
4: Best ← SmallnessTournament(P, t2 )
5: for i from 2 to t1 do
6: Next ← SmallnessTournament(P, t2 )
7: if Fitness(Next) > Fitness(Best) then
8: Best ← Next
9: return Best
10: procedure SmallnessTournament(P, t2 )

12: for i from 2 to t2 do
14: if Next is smaller than Best then
15: Best ← Next
16: return Best
Now we have two tournament sizes: t1 and t2 . Assuming we keep with GP tradition and have
t1 = 7, a good setting for t2 appears to be 2. Actually, the best value is around 1.4: remember that
for values of t less than 2, with probability t − 1.0, we do a tournament selection of size t = 2, else
we select an individual at random. You could do it the other way around too, of course: pick by
fitness first, the by smallness. There are lots of choices for parsimony pressure, but these two give
you the general idea (and are based on tournament selection to boot!).
92
5 Parallel Methods
Metaheuristics can be expensive. It’s not uncommon, for example, to see over well over 100,000
assessments per run in Genetic Programming (say, a population size of 2000, run for 50 generations).
And assessments can take a while to run: perhaps they’re simulations, or complex chemical
structure analyses. As a result, parallel methods are enticing.
I believe this is the strongest argument for parallel methods. But there are those in the com-
munity who argue that certain parallel methods (notably Island Models, discussed in Section 5.2)
have a positive effect on the optimization process itself. For example, Zbigniew Skolicki73 identi-
fied fitness functions where parallel methods are better than a single evolutionary computation
population even if you discount the speedup enjoyed by multiple machines.
Lots of stochastic optimization techniques can be parallelized: but some perhaps more easily
than others. Single-state methods (hill-climbing, simulated annealing, tabu search, etc.) are
parallelizable but, in my opinion, only in awkward ways. Perhaps the most readily parallelizable
methods are the population methods, since they already deal with many simultaneous candidate
solutions which need to all be assessed. The five biggest ways to parallelize:74
• Do separate runs in parallel.
• Do one run which splits the fitness assessment task (and possibly also the breeding and
initialization tasks) among multiple threads on the same machine.
• Do separate runs in parallel which occasionally hand high-quality individuals to one another
(spreading the goodness). These are known as Island Models.
• Do one run which, when it needs to assess the fitness of individuals, farms them out to remote
machines. This is known as Master-Slave or Client-Server Fitness Assessment.
• Do one run with a selection procedure which presumes that individuals are spread out in a
parallel array on a vector computer (called spatially embedded or fine-grained models).
These five can also be mixed in quite a lot of ways. There’s no reason you can’t do an island
model where each island does master-slave fitness assessment, for example.
Thread Pools Several of the following algorithms assume that the threads have spawned them-
selves and have inserted themselves into a thread pool from which we may draw them and tell them
to start; when they finish they reenter the thread pool again.
73 Zbigniew Skolicki, 2007, An Analysis of Island Models in Evolutionary Computation, Ph.D. thesis, George Mason
University, Fairfax, Virginia.

74 Though many people have contributed to the literature on methods like these, John Grefenstette’s 1981 technical
report on the subject was unusually prescient. His algorithm A described a multiple-thread or Master-Slave method;
algorithms B and C described Asynchronous Evolution (a Master-Slave variant described later), and algorithm D
described Island Models. From John Grefenstette, 1981, Parallel adaptive algorithms for function optimization, Technical
Report CS-81-19, Computer Science Department, Vanderbilt Univesity.
93
Algorithm 65 Thread Pool Functions
1: global l ← lock for the pool
2: global T ← { } empty pool of tuples !t = $ tlock , tdata % where tlock is a lock and tdata is any object
3: procedure InsertMyselfAndWait( )
4: Acquire lock l
5: !t ← new tuple $tlock , tdata % " tlock is a new lock. tdata can be anything for now
6: T ← T ∪ {t}
7: Acquire lock tlock
8: Notify threads waiting on l
9: Wait on tlock " This releases both locks, waits to be notified on tlock , then reacquires the locks
10: o ← copy of tdata " At this point tdata was set in TellThreadToStart(...)
11: Release lock tlock
12: Release lock l
13: return o
14: procedure ThreadIsInserted( )

15: Acquire lock l
16: if T is empty then
17: Release lock l
18: return false
19: else
20: Release lock l
21: return true
22: procedure TellThreadToStart(information o)

23: Acquire lock l
24: while T is empty do
25: Wait on l " This releases the lock, waits to be notified on l, then reacquires the lock
26: t ← arbitrary tuple in T
27: T ← T − {t}
28: Acquire lock tlock
29: tdata ← copy of o
30: Notify threads waiting on tlock
31: Release lock tlock
32: Release lock l
33: return t
34: procedure WaitForAllThreads(number of threads n)

35: Acquire lock l
36: while || T || < n do
37: Wait on l
38: Release lock l
94
This can be complicated and hard to debug. The thread pool in turn requires the ability to:
• Spawn threads.
• Acquire and release locks on a per-thread basis. If a thread tries to acquire a lock already
acquired by someone else, the thread pauses until the lock is released to them.
• Have the ability to wait on a lock, meaning to release the lock to others and pause until
someone else has notified you regarding that lock.
• Have the ability to notify threads waiting on a given lock. This causes the threads to one-by-
one reacquire the lock, unpause themselves, and go on their merry way.
Basically every threading library provides this functionality. Using this, we can spawn as many
threads as we want initially, and direct them to call InsertMyselfAndWait to get the next individual
they’re supposed to work on. This is pretty standard stuff but it’s a bit complex.
5.1 Multiple Threads

To assess a population, we could hand off individuals one by one to threads as they come available.
When a thread finishes an individual, it is made available for another one.
Algorithm 66 Fine-Grained Parallel Fitness Assessment
1: P ← current population

3: TellThreadToStart({ Pi }) " If total threads ≥ || P||, one will always be available
4: WaitForAllThreads()
5: return P
This requires use of a thread pool. A simpler approach, which requires no locks, just breaks the
population into chunks and hands each chunk to a separate newly spawned thread. Then at the
end we just gather up all the threads.
Algorithm 67 Simple Parallel Fitness Assessment
1: P ← population { P1 , ..., Pl }
2: T ← set of threads { T1 , ...Tn }
4: a ← 0l/n1 × (i − 1) + 1 " Figure out the lower (a) and upper (b) boundaries for chunk i
5: if i = n then
6: b←l
7: else
8: b ← 0l/n1 × i
9: Spawn Ti and tell it to Assess individuals Pa through Pb
11: Wait for Ti to exit
12: return P
95
Here we just need to be able to spawn threads and wait for them to finish (both standard
functions in a thread package). You can “wait” for the threads to finish simply by “joining” with
them (a standard function in parallel toolkits). The downside of this is that it’s often the case that
one chunk happens to have all the slow-to-assess individuals (if such things exist), and the other
threads will wind up sitting around idling waiting for that last thread to finish up.
It’s possible to do the same tricks to parallelize population initialization, though it’s rare for
it to take long to initialize a population. Still, Algorithm 66 could be used except that instead of
evaluating an existing individual Pi , each thread is told to create an individual and insert it into slot
i. Similarly, Algorithm 67 could be used for initialization, where instead of doing Assess(Pa ...Pb ), we
tell each thread to initialize b − a + 1 individuals and stick them in slots a...b.
Likewise, you could do the same tricks for breeding, but it can be complicated by choice of
selection procedure if certain things need to be done offline. Tournament Selection works nicely,
as it doesn’t require any beforehand, offline stuff. Note that we’re no longer dividing up the
population P into chunks, but rather dividing up the next-generation population Q into chunks:
Algorithm 68 Simple Parallel Genetic Algorithm-style Breeding

2: T ← set of threads { T1 , ...Tn }
3: l ← desired size of new population
4: !q ← empty array $q1 , ..., ql % " Will hold the newly bred individuals
6: a ← 0l/n1 × (i − 1) + 1
7: if i = n then
8: b←l
9: else
10: b ← 0l/n1 × i
11: Spawn Ti and tell it to Breed individuals into slots q a through qb
13: Wait for Ti to exit
14: return !q converted into a population
Algorithm 69 Fine-Grained Parallel Genetic Algorithm-style Breeding

2: l ← desired size of new population " Presumably even in length
3: !q ← empty array $q1 , ..., ql %

4: for i from 1 to l by 2 do " We jump by 2 to allow crossover to make two kids
5: TellThreadToStart({!q, i, i + 1}) " The thread breeds from P into slots qi and qi+1
6: WaitForAllThreads(total number of threads)
7: return !q converted into a population
96
G H I
A C A B C A
B
B C D F D E F D A D E
C
E I G H I G
A B C
Figure 36 Fully-Connected, 2-D Toroidal Grid, and Injection Island Model topologies.
The reason these algorithms work is that breeding is a copy-forward procedure: we select indi-
viduals in P, copy them, then modify the copies. So we never need to bother locking on individuals
in P. Other procedures might require more complexity than this.
5.2 Island Models

An island model is a group of simultaneous evolutionary optimization processes which occasionally
send individuals to one another to help spread news of newly-discovered fit areas of the space.
Some island models send the fittest individuals they have, or use a selection operation to pick them.
Others send random individuals (which is less exploitative).
Island models are primarily meant to take advantage of computational resources, commonly by
loading each of n computers with one process each. But they have another feature: because the
population is broken into separate subpopulations (sometimes called demes), it can take longer
for fit solutions to take over the whole population, which makes the whole system more diverse
and explorative. Another argument made in favor of island models is that if you have a fitness
function which can be broken into different components, island models might theoretically help
those different components be developed in separate populations.
To set up an island model you need to define an island topology. Which islands will send
individuals to which other islands? There are three common ones, shown in Figure 36. A fully-
connected topology has every island talking with every other island. A toroidal grid topology lays
out the islands in, well, an n-dimensional toroidal grid. In a fully-connected topology, spreading an
individual from one island to any other one requires a single hop, but in a grid topology, it can take
a while for an individual to wander through various populations to get to a distant population.
Thus we might expect a grid to promote diversity more.
One more topology: an injection model is a feed-forward structure. Sometimes this is used to
assist the fitness function. Let’s say you are looking for a robot soccer player. The EA can’t come up
with everything at one time, so you’ve constructed a pipeline to help things by constantly upping
the ante. In an early island, the fitness function is: how well can you kick? Individuals then migrate
to a later island where the fitness function has changed to: how well can you pass the ball? Further
individuals might migrate to another island where the objective is to keep the ball away from an
opponent by kicking carefully between teammates. And so on.
Once you’ve settled on a topology, you’ll need to modify the EA. Here’s how I do it:
97
Algorithm 70 An Abstract Generational Evolutionary Algorithm With Island Model Messaging
2: Best ← !
3: repeat
4: AssessFitness(P)
5: Send copies of some individuals from P to mailboxes of neighboring islands
8: Best ← Pi
9: M ← extract and return all contents of my mailbox
10: P ← Join(P, M, Breed(P)) " It’s possible you may breed individuals but never use them
12: return Best
We’ve augmented the abstract EA with a mailbox metaphor: each island has a mailbox to
which other islands may send individuals. Islands can, at their leisure, extract and return the
individuals presently in their mailboxes. We’ve added three gizmos to the abstract algorithm using
this metaphor: sending individuals to neighboring islands, receiving individuals that have shown
up in our mailbox from neighboring islands, then joining those individuals with the population
along with the others (notice that Join now takes three arguments). If you’re doing an injection
model, individuals arriving at your mailbox may need to have their fitnesses reassessed given your
own island-specific fitness function.
It’s possible that your particular approach will require a synchronous algorithm where the
islands all wait until everyone has caught up before sending individuals to one another. But in
most cases an asynchronous mechanism makes better use of network resources. Here individuals
just get sent whenever, and pile up at the doorstep of the receiving island until it’s ready to take
them in. This allows some islands to be slower than others. Of course, in that situation you’ll need
to decide what to do if a mailbox is overflowing.
Another issue affecting network throughput is the number and kind of connections in your
topology. Which machines are connected? How often do they send individuals to each other, and
when? How many individuals get sent? Lots of connections, or poorly considered topologies, can
stress parts of your network. In my department we have a cluster with two networks, each with
its own router. The two routers are then connected together on a fast channel but not fast enough.
Here, I’d probably want a network configuration where nodes on the same router talk to each
other much more often. Additionally I might configure things so that each island only sends out
individuals every m generations, and those islands stagger when they send individuals. When
doing layouts like this, think about how to maximize throughput.
In Section 5.1 we went in detail about how to handle the locking etc. to do the parallel model.
For island models we won’t do this detail, but it’s not super complex. You connect to your neighbor
islands via sockets, then either use the UNIX select() function, or spawn a separate thread for each
socket. Let’s presume the latter. The thread loops, reading from the socket, locking on the mailbox,
adding to the mailbox, then releasing the lock. In the main EA, to get the current contents of the
mailbox and clear it, you just need to acquire the lock first, then release it afterwards. Sending
individuals to neighbors is just a matter of writing to the remote socket (no threads involved).
98
5.3 Master-Slave Fitness Assessment
This is the most common form of parallel metaheuristics, and it’s also
among the most straightforward. The machines at your disposal are Slave 1
75
divided up into a master and some n slaves. When you need to assess
an individual, you send it off to a slave to do the job. This becomes more Slave 2
useful as the time cost of assessing the fitness of an individual becomes
Master
high. And for a great many optimization tasks of interest nowadays,
fitness assessment time is so long that it’s far and away the dominant Slave 3
factor in the optimization process.

So how do you set up a master-slave arrangement? Basically the same Slave 4
way you do multithreaded fitness assessment (Algorithms 65 and 66 — I
wouldn’t bother with Algorithm 67). The only real difference is that each
Figure 37 Master-slave fit-
thread registered in the thread pool is assigned to talk to a particular ness assessment.
slave over the socket. Instead of assessing the individual (or individuals)
directly in the thread, the thread sends it over a socket to get assessed remotely. You’ll need to
modify Algorithm 66 to allow a thread to receive multiple individuals at once, like this:
Algorithm 71 Fine-Grained Master-Side Fitness Assessment

1: P ← current population { P1 , ..., Pl }
2: n ← number of individuals to send to the slave at one time
3: for i from 1 to l by n do
4: TellThreadToStart({ Pi , ..., Pmin(i+n−1,l ) }) " The thread will ship them off to the remote slave
5: WaitForAllThreads()
6: return P
This approach is also nice because it’s relatively graceful with slow slaves, and with variable-
length fitness assessment times, and also allows new slaves to show up. To robustly handle slaves
which disappear while assessing fitness requires more complexity which we won’t bother with
here (though it’s important!).
When is a master-slave process useful? It’s all down to how much network capacity and speed
you have. A master-slave approach becomes useful when you have enough capacity and speed that
the time spent shipping an individual to a remote site and back is less than just assessing the fitness
on your own processor. There are a couple of tricks you can employ to maximize throughput. First,
compress your individuals when sending out on the socket, particularly if they’re GP or other
individuals with a lot of redundancy. Second, in many cases you don’t need an individual sent
back to you from the slave: you may just need its fitness (it depends, as we’ll see in a moment).
Third, many networks send in packets that incur a constant overhead: you’d like to load up that
packet if you can. Sending a chunk of individuals to a slave may fill the packet nicely, but sending
a single individual can waste a lot of overhead.
Here’s how I see master-slave versus island models. If your throughput is low and your fitness
assessment time is very short, you might pick an island model. Or you might pick an island model
if you wish to optimize using a very large population. Otherwise I’d pick master-slave.
75 Or if you like, a client and servers, or is it a server and clients?
99
You could mix the two as well. For example, you could have a bunch of islands, with each
island assigned its own set of fitness assessment slaves. Or you could try what I call Opportunistic
Evolution,76 which is particularly good for grid computing scenarios. Here, you have a bunch of
slaves as usual, and send some n individuals to each slave at a time. Each slave is also given a
certain large time interval, big enough to justify shipping the individuals off to it. When a slave has
finished assessing its individuals, and there’s more time left in the interval, the slave does a little
optimization (perhaps hill-climbing or evolutionary computation) with its n individuals as a mini
population. When time is up, the slave returns the revised mini-population to the master rather
than the original individuals. (Note that to do this you must send individuals back to the master,
not just fitnesses).
You can also mix Master-Slave Fitness assessment with a Steady-State Genetic Algorithm in
a fairly elegant way, a notion called Asynchronous Evolution. Whenever a slave is ready to
receive individuals, we select and breed individuals right there and ship them off to the slave.
Asynchronous Evolution doesn’t wait for slaves to all complete — it’s asynchronous — but rather
whenever a slave has finished (taking as long as it likes), its individuals get Joined into the
population. Different slaves can finish at different times. This approach tolerates an extremely
wide variance in fitness assessment time: of course, long-evaluating Individuals may be at a
disadvantage because they can’t breed as often.
Asynchronous Evolution relies on a threadsafe collection (multiset). Whenever a thread receives
completed individuals from its remote slave, the thread inserts them into the collection using the
AddToCollection(...) function. The Asynchronous Evolution algorithm itself polls for new completed
individuals by repeatedly calling the RetrieveAllFromColection(...) function. Trivially:
Algorithm 72 Threadsafe Collection Functions

1: global S ← {}
2: global l ← lock for the collection S
3: procedure AddToCollection(T) " T is a set of things

4: Acquire lock l
5: S ← S∪T
6: Release lock l
7: procedure RetrieveAllFromCollection( )
8: Acquire lock l
9: T←S
10: S ← {}
11: Release lock l
12: return T
Given the threadsafe collection, Asynchronous Evolution proceeds as follows.

76 This notion was first suggested in a technical report:
Ricardo Bianchini and Christopher Brown, 1993, Parallel genetic
algorithms on distributed-memory architectures, Revised Version 436, Computer Science Department, University of
Rochester, Rochester, NY 14627.
The term Opportunistic Evolution was coined by Steven Armentrout for a paper we wrote: Keith Sullivan, Sean Luke,
Curt Larock, Sean Cier, and Steven Armentrout, 2008, Opportunistic evolution: efficient evolutionary computation
on large-scale computational grids, in GECCO ’08: Proceedings of the 2008 GECCO Conference Companion on Genetic and
Evolutionary Computation, pages 2227–2232, ACM, New York, NY, USA.
100
Algorithm 73 Asynchronous Evolution
1: P ← {}
2: n ← number of individuals to send to a slave at one time
4: Best ← !
5: repeat
6: if ThreadIsInserted( ) = true then " Check for processors with nothing to do
7: if || P|| < popsize then " Still initializing population
8: Q ← n new random individuals
9: else " Steady State
10: Q ← {}
11: for i from 1 to n by 2 do " Obviously we could do some other kind of breeding
15: Q ← Q ∪ { Mutate(Ca ), Mutate(Cb ) }
16: TellThreadToStart({ Q1 , ..., Qn })
17: M ← RetrieveAllFromCollection( ) " Get all individuals who have completed fitness assessment
18: for each individual Mi ∈ M do
19: if Best = ! or Fitness(Mi ) > Fitness(Best) then
20: Best ← Mi
21: if || P|| = popsize then " Steady State
23: P ← P − { Pd }
24: P ← P ∪ { Mi }
25: if ThreadIsInserted( ) = false and M is empty then
26: Pause for a little while " Nothing’s going in or out: give your CPU a break
28: return Best
5.4 Spatially Embedded Models

1 2 3 4 5
A spatially embedded model adds to the population a notion of
physical locations of individuals in the population. For example, the 6 7 8 9 10
population may be laid out in a 3D grid, or a 1D ring, and each
11 12 13 14 15
individual occupies a certain point in that space. Figure 38 shows
individuals laid out in a 2D grid. 16 17 18 19 20
Such models are mostly used to maintain diversity in the popu-
lation, and so promote exploration. Individuals are only allowed to 21 22 23 24 25
breed with “nearby” individuals, so a highly fit individual cannot
spread as rapidly through a population as it could if there were no Figure 38 A spatially embedded
population of individuals in a 2D
breeding constraints. Notionally regions of the population might grid. Individuals 1...25 shown.
develop their own cultures, so to speak, similar to Island Models.
101
Spatially embedded models are parallelizable in the multi-threaded sense. But if you have a
vector processor, that is, a machine which performs many identical simultaneous operations at one
time, there exist techniques for doing all the breeding and selection in parallel on the processor.
Nowadays the most common vector processor in your possession is your graphics processing unit
(GPU). Assuming we stick with the multithreaded example, we can easily do multi-threaded fitness
assessment and multi-threaded breeding. We just need a modification of the breeding procedure.
Algorithm 74 Spatial Breeding

1: P ← current population, organized spatially
2: Q ← new population, organized spatially in same way as P

3: for each individual Pi ∈ P do " This can be done in parallel threads as before
4: N ← Neighbors(Pi , P)
5: Parent Na ← SelectWithReplacement(N)
6: Parent Nb ← SelectWithReplacement(N)
7: Children Ca , Cb ← Crossover(Copy(Na ), Copy(Nb ))
8: Qi ← Mutate(Ca ) " Cb is discarded. Ca goes directly to spatial slot i in Q
9: return Q
Or if you like, we could just do mutation, no crossover. The important part is that we are
specifically replacing individuals, in each slot, with children bred from neighbors in that area.
To replace an individual Pi , selection is performed not from
the whole population but rather on a subset N of neighbors of
1 2 3 4 5
Pi . It’s up to you to define the neighborhood function. You could
define neighbors of Pi as the individuals which reside within a box 6 7 8 9 10
centered at Pi ’s location and of m size in each dimension. Or you
could build up N by repeatedly doing random walks starting at 11 12 13 14 15
Pi ; each time you finish the walk, you add the final individual to 16 17 18 19 20
N. Individuals selected this way are chosen roughly according to
a gaussian-ish distribution centered Pi . The longer the walk, the 21 22 23 24 25
larger the neighborhood. For example:
Figure 39 A random walk in the
Algorithm 75 Random Walk Selection grid, starting at individual 12.
2: r ← chosen random walk length
3: Pi ← Start Individual
4: !l ← location $l1 ...ln % of Pi in the space

5: for r times do
6: repeat
7: d ← random integer chosen uniformly between 1 and n inclusive
8: j ← either a 1 or -1, chosen at random.
9: until ld + j is an acceptable value for dimension d in the space
10: ld ← ld + j " Wrap-around as necessary if in a toroidal environment
11: !
return the individual located at l in the space
102
6 Coevolution
Coevolution can mean a lot of things in biology, but the classic notion is one of different species
acting as foils against one another and causing one another to adapt. For example, a particular
fungus might develop an antibiotic to kill a particular bacterium, and the bacterium then adapts to
not be harmed by that antibiotic, forcing the fungus to construct a tougher antibiotic. Coevolution
could also include symbiotic relationships: leafcutter ants and the fungus they farm for food, both
co-adapting to work better as a team, so to speak.
Coevolution is also a common framework in metaheuristics, and is usually applied to
population-based optimization methods. In optimization, coevolution generally refers to situ-
ations where the fitnesses of individuals in a population are affected by the presence of other
individuals in the population(s). More formally, we’d say that a system exhibits coevolutionary
properties if the question of whether individual A is superior to individual B is dependent on the
presence or absence of some other individual or individuals C in the population. For example, let’s
say that the fitness of an individual is based on competing with other individuals in the population
in Tic-Tac-Toe. A usually wins more games than B does, so A has a higher fitness. But whenever A
plays C, A loses badly. Curiously C always loses to B! So if C is in the population and playing, then
A’s fitness may drop to less than B’s. The fitness is context-sensitive.
Such scenarios lead to certain odd pathological conditions: and indeed there are a lot of open
issues in coevolution dealing with ironing out these conditions and getting coevolutionary systems
to behave more like regular optimization techniques. This is a well-motivated area of research
because coevolutionary methods are promising in important ways! Coevolution can allow a system
to gracefully ramp up in difficulty, to provide diversity in the system, to discover not just high-
quality but robust solutions, and to solve complex, high-dimensional problems by breaking them
along semi-decomposable lines.77
Here are the main kinds of coevolutionary techniques:78
• 1-Population Competitive Coevolution Individuals in a single population base their fit-

ness on games they play against one another. Commonly used to evolve good competitive
strategies (for checkers or soccer, for example).
• 2-Population Competitive Coevolution The population is broken into two subpopula-
tions (or demes). The fitness of an individual in subpopulation 1 is based on how many
individuals in subpopulation 2 it is able to defeat in some competition (and vice versa). Com-
monly subpopulation 1 contains the candidate solutions of interest to us, and subpopulation
2 contains foils meant to test them. Usually used to force subpopulation 1 to find robust
candidate solutions despite everything subpopulation 2 can throw at them.
77 Coevolutionary algorithms are sometimes motivated because they’re more “biological”, if that matters to you from
an engineering perspective (it oughtn’t). I recall a biologist remarking that coevolutionary algorithms should have
instead been called evolutionary algorithms, because the algorithms we call evolutionary algorithms really have little to do
with evolution. And that’s basically right. Evolutionary algorithms are more like dog breeding: you select and breed the
dogs, er, algorithms based on assessments of them independent of one another. That is, you’re doing a form of artificial
directed selection. But in real natural selection, individuals survive based on the makeup of their particular populations
(helping and/or competing with them) and the presence of certain predators or prey. It’s context-sensitive.
78 The terms competitive and cooperative coevolution come from similar notions in game theory. But a dangerous faction
of theorists prefers to call them compositional if the objective is to put pieces together to form a complete solution (like
N-Population Cooperative Coevolution) and test-based, if the objective is for one population to provide a foil for the
other population (as is usually done in 2-Population Competitive Coevolution).
103
• N-Population Cooperative Coevolution The problem to be solved is divided into n sub-
problems: for example, if the problem is to find soccer strategies for a team of n robots, it’s
divided into subproblems, each of which is to find a strategy for one of the robots. The task
of finding each of these subsolutions is given to each of n subpopulations. The fitness of an
individual (in this case, a robot strategy) is assessed by selecting individuals from the other
subpopulations, grouping them with the individual to form a complete n-sized solution (a
complete soccer robot team), and determining the fitness of that solution. Commonly used to
reduce the high dimensionality of big problems by decomposing them into multiple simpler
problems.
• Diversity Maintenance (Niching) Individuals in a single population are forced to spread

portions of their fitness to one another, or to be grouped into competitive niches (or species),
etc., to impose diversity on the population.
These are the main groups. But there are many other variations on these themes.79 For example,
in what I call Cooperative-Competitive Coevolution, a solution is composed of multiple subpieces
as in N-Population Cooperative Coevolution. But the solution is tested by playing it against other
such solutions in a game, as in 1-Population Competitive Coevolution. For example, imagine that
we’re looking for soccer teams consisting of a goalie, forwards, midfielders, and defenders. We
have four subpopulations (goalies, forwards, etc.). An individual (a goalie, say) is assessed by
selecting from these subpopulations to form a team which includes that individual. Then we select
from the subpopulations to form an opposing team, and play a game of soccer.
Because coevolution is most commonly seen in population-based methods (notably Evolution-
ary Computation), the algorithms described here are in that context. But there is no reason why it
can’t be applied in limited form to single-state metaheuristics. For example, to use a hill-climber,
just define your “population” as your current two individuals (parent and child). To avoid the
evolution stigma, we might instead call such algorithms co-adaptive rather than coevolved.
What does Fitness Mean Now? The fitness of individuals in a coevolutionary system is a relative
fitness based on how they performed in the context of individuals in the same optimization process.
For example, if the individuals were competing against one another in a game, an individual in
generation 0 might have a decent fitness (because the other individuals in generation 0 are awful),
but if it was magically teleported to generation 100 its fitness would be terrible (because they’ve
improved). We no longer have an absolute fitness.
The appearance of relative fitness creates two gotchas. First, it mucks with the dynamics of
selection and breeding, which can result in problematic operation of the system, as we’ll see in a
bit. Second, it’s a big problem if we want to assess how well the algorithm is doing. Previously
we could just sample individuals out of each generation and see their fitnesses going up and up.
But now it’s more likely that the individuals will be improving, but the fitnesses will be staying
roughly the same because their opponents are improving as well.
79 Certain other techniques discussed later exhibit coevolutionary traits (and likely pathologies). Ant Colony Opti-
mization (Section 8.3) and Michigan-Approach Learning Classifier Systems (Section 10.4) both have features similar
to the 1-Population methods above. And Univariate Estimation of Distribution Algorithms (Section 9.2.1) turn out to
have a very close theoretical relationship with N-Population Cooperative Coevolution. Island Models, discussed earlier
in Section 5.2, also use subpopulations, and also have certain diversity-maintenance features, and so exhibit certain
relationships with coevolutionary models.
104
This brings up two common uses of fitness: an internal fitness is a measure used by the
optimization system to determine selection. An external fitness is a measure used to examine the
quality of an individual in order to gauge the progress of the algorithm. As we’ve seen, internal
fitnesses can be either relative or absolute, but we’d like the external fitness to be absolute. We will
distinguish between these using AssessInternalFitness(...) and AssessExternalFitness(...).
As it’s relative, an internal fitness may change based on the tests performed in the context of
other individuals in the population. Thus you may need to re-assess the fitness of all individuals
each time around.
Tests Sometimes fitness assessments are done by doing a collection of tests on your individual;
commonly the fitness is the sum or average of the test results. This might be because you have a
fixed set of test cases (for example the Genetic Programming / Symbolic Regression example in
Section 4.3 is often done by testing each individual against some m test cases). It can also be because
you have a very large (often infinite) set of possible situations for which you’d like your individual
to be optimal, or at least pretty good. We call an individual which is good in lots of situations, even
if not optimal anywhere in particular, a robust individual. Such tests thus would sample a lot of
places in the situation space. You might also perform multiple tests with an individual to find a
location in the space where it is particularly impressive, even if it’s not impressive everywhere.
Last, multiple tests might be important in order to weed out noise.
Multiple-test fitness assessment shows up over and over in coevolution. It’s a natural fit because
if you’re testing an individual against other individuals, you might as well test against a bunch
of them to get a good sample. Each of the methods discussed here will employ fitness tests for
different purposes, so the algorithms usually will have two stages to fitness assessment: first, gather
some m tests for each individual (using Test(...)) in the context of other individuals, then assess the
fitness of the individual based on those tests.
6.1 1-Population Competitive Coevolution

1-Population Competitive Coevolution is mostly used for optimizing candidate solutions designed
to compete in some kind of game. For example, Kumar Chellapilla and David Fogel80 used 1-
Population Competitive Coevolution to search for good-quality checkers players.81 I also used this
kind of coevolution to search for robot soccer team strategies. The idea is simple: each individual’s
fitness is assessed by playing that individual against other individuals in the population in the
game of interest (checkers, soccer, whatnot).
The intuition behind this idea is to improve the learning gradient of the search space. Imagine
that you’re trying to find a good poker player. One way to do it would be to build a really good
“guru” poker player by hand, then assess the fitness of individuals based on how many hands, out
of n, in which they beat the guru. The problem is that the vast majority of random individuals are
awful: they lose every single hand to the guru. Early on your population will consist entirely of
these individuals and so the optimization system wouldn’t be able to tell which of the individuals
were better than the others. The quality function would be flat until you find an individual who’s
sometimes able to beat the guru, which is rare. If you are so lucky, then it’s easier going from there.
80 Son of Larry Fogel, of Evolutionary Programming fame. See Section 3.1.1.
81 More specifically: they were looking for neural networks which performed the board evaluation function to be
attached to standard α-β game tree algorithms for checkers. See also Footnote 83. The paper was Kumar Chellapila and
David Fogel, 2001, Evolving an expert checkers playing program without using human expertise, IEEE Transactions on
105
Figure 40 shows this situation. Basically your system can’t get started because there’s no way to
differentiate among all the initial individuals because they’re so bad against the guru. Until you get
players who can start beating the guru sometimes, it’s essentially a needle-in-a-haystack scenario.
You could fix this by coming up with a special way of assessing fitness among the individuals who
always lose against the guru: how badly did they lose? Did they do something smart somewhere?
Etc. But this is often quite complicated to do. Alternatively you could create a panel of hand-built
custom players, from very simple, stupid ones, all the way up to the guru. Individuals would be
assessed based on how many of the custom players they
beat. This would present a more gentle hill for the system (AverageFitness
Hands Won
to climb up. Even so, if you’re smart enough to be able to Against the Guru)
construct a wide range of poker playing programs to test 1.5
your individuals, why in the world would you be using

stochastic optimization to find a good poker player? Don’t 1
Without any gradient to lead to it,
you already know how to do that? this cliff will only be discovered by chance.
{
Essentially a needle in a haystack.
An alternative route (1-Population Competitive Coevo- 0.5
lution) is to have your individuals play each other. Ini-

tially, an individual is bad, but the players he’s being tested 0
Most individuals always lose
against are bad too. Someone’s got to win the poker hand,
so your players won’t have all zero fitness. Your optimiza- Figure 40 A tough learning gradient when
tion algorithm can tell which ones are (marginally) better. playing poker with a guru.
Later on, as the individuals improve, so do the players against
which they are being tested. As your individuals get better, the problem naturally gets harder. We’ve
created an automatically self-adjusting learning gradient.
Handling external fitness can be a problem. Some ways to compute an external fitness:
• Test against a guru or against a panel of hand-created players.
• Test against a sample of players from previous generations (assuming they’re not as good).
• Test against some external system in the real world.
The first option has the same issues as was discussed before. The second option makes a
big assumption: that your later players are better than your earlier individuals. Essentially your
external fitness wouldn’t be an absolute measure but relative to other players, which could create
some odd results if the system’s not stable.
The third option is what Chellapilla and Fogel did for their checkers player. To test the progress
of their system, they occasionally collected an individual from the current generation and played
it on MSN Gaming Zone82 against real people in the site’s checkers room, eventually garnering a
checkers ranking. The system was improving if the ranking was going up.83 Chellapilla and Fogel
82 http://games.msn.com/
83 This leads to a notorious anecdote. At the time, the checkers room of MSN Gaming Zone consisted largely of
teenage and twentysomething males. When the evolved checkers player was losing against these people, they’d play
to the end. But when it was beating them, they’d drop the game. Thus Chellapilla and Fogel couldn’t collect enough
positive samples to assess their program, just negative samples. This led them to adopt the fictional online persona of a
24-year-old, beautiful female PhD student who went by the username of Blondie24. The MSN Gaming Zone players,
thinking they were playing against a woman, would play to the very end regardless of whether they were winning
or not. Fogel later wrote a book about the experience: David Fogel, 2001, Blondie24: Playing at the Edge of AI, Morgan
Kauffman. See also Footnote 81.
106
also used the “guru” option for their external fitness, by playing against Chinook,84 a well-regarded
checkers program, and one which can be adjusted to play at different levels of difficulty.
This external fitness metric is the primary thing that distinguishes the abstract algorithm from
the abstract version of a traditional evolutionary algorithm. Here’s the abstract algorithm:
Algorithm 76 An Abstract Generational 1-Population Competitive Coevolutionary Algorithm

2: Best ← !
3: repeat
4: AssessInternalFitness(P) " Used for Selection (in Breeding)
5: AssessExternalFitness(P) " Used to determine algorithm progress and Best
7: if Best = ! or ExternalFitness(Pi ) > ExternalFitness(Best) then
8: Best ← Pi
11: return Best
1-Population Competitive Coevolution could also be cast in an asynchronous Steady State

form. As part of their work with N EAT, Ken Stanley and Risto Miikkulainen did a project involving
the optimization of robot game agents in a free-for-all 2D shoot-em-up.85 The entire population
participated at the same time in the game, shooting at each other. Those individuals with better
skills lasted longer; the less fit individuals, once shot, would be removed from the game and
replaced with children newly bred from the remaining population. An approach like this is fairly
rare, but intriguing from a biological or social modeling perspective. Perhaps a method like this
might be helpful for, say, for modeling competing restaurants in a city, or competing gangs, etc.
6.1.1 Relative Internal Fitness Assessment

There are a variety of ways that one might assess the fitness of individuals by pitting them against
other individuals in the population. The primary issue that you’re going to face is the number
of tests (evaluations) involved in order to assess the fitness of the population. Many games,
competitions, or similar things are stochastic: you might win one time, lose another. For example,
in poker, you might need to play a bunch of hands to determine which player is doing better.
Even if your game isn’t stochastic, it’s also common that you’re going to run into a major
pathology in 1-Population Competitive coevolution: the presence of cycles among game players.
That is, Individual A beats Individual B, and B beats C, but C beats A. In that case, who is better?
84 http://www.cs.ualberta.ca/∼chinook/ Chinook was also the first program to win a the world championship in a
nontrivial game. Marion Tinsley (the greatest human checkers player ever) wanted to play Chinook for the championship
after Chinook started winning competitions. But the American and English checkers associations refused. So Tinsley
forced their hand by resigning his title. They gave in, he got to play Chinook, and he won 4 to 2 with 33 ties. On the
rematch four years later, after 6 ties, Tinsley withdrew because of stomach pains; and died soon thereafter of pancreatic
cancer. So Chinook won, but sadly it did so by default. It’s since improved to the point that Chinook likely cannot be
beaten by any man or machine. But who knows if Tinsley would have won?
85 The game environment is called N ERO : Neuro-Evolving Robotic Operatives. http://nerogame.org/ Coevolution
with N EAT was discussed in Kenneth O. Stanley and Risto Miikkulainen, 2004, Competitive coevolution through
evolutionary complexification, Journal of Artificial Intelligence Research, 21, 63–100.
107
Worse still, perhaps A beats B, but B beats far more individuals than A beats. Who’s better in that
situation? It depends on what you’re looking for. Often the case is that you’re looking for the
individual which wins against the most players; or against the most “good” players, or wins by the
most total points on average. In such cases it’s common to require multiple tests to identify which
individuals are really the better ones.
The tradeoff here is how many tests you need to run. If you have a fixed budget of tests, you
can spend them on more precisely determining fitness; or you can spend them on searching further.
How to spend that budget is yet another parameter to deal with. Different ways of assessing the
fitness of individuals lie at different points along this tradeoff spectrum. One simple approach is to
pair off all the players, have the pairs play each other, and use the results as fitness. We introduce a
Test function which tests two individuals (plays them against each other) and stores the results,
which are then used by the fitness assessment procedure. The simple approach is:
Algorithm 77 Pairwise Relative Fitness Assessment

1: P ← population
2: Q ← P, shuffled randomly " To shuffle an array randomly, see Algorithm 26

3: for i from 1 to || Q|| by 2 do " We assume || Q|| is even
4: Test(Qi , Qi+1 )
5: AssessFitness(Qi ) " Using the results of the Test
6: AssessFitness(Qi+1 ) " Using the results of the Test
7: return P
This has the advantage of only requiring || P||/2 tests to assess the fitness of individuals in
the population. But each individual only gets to be tested against one other individual, which
is probably very noisy. An alternative is to test individuals against every other individual in the
population. The fitness is then based on the various tests the individual received, like so:
Algorithm 78 Complete Relative Fitness Assessment

1: P ← population
2: for i from 1 to || P|| do

3: for j from i + 1 to || P|| do
4: Test(Pi , Pj )
5: AssessFitness(Pi ) " Using the results of all Tests involving Pi
6: return P
|| P||×(|| P||−1)
This involves each individual in || P|| − 1 tests, but as a result it requires 2 tests all
told! That’s a lot of tests. Is the tradeoff worth it? We can strike a middle ground between these two
by testing each individual against some k other individuals in the population, chosen at random.
Some individuals will ultimately get tested more than k times, but each individual will get at least k
tests to assess its fitness. The algorithm looks something like this:
108
Algorithm 79 K-fold Relative Fitness Assessment
1: P ← population
2: k ← desired minimum number of tests per individual

4: Q ← k unique individuals chosen at random from P, other than Pi
6: Test(Pi , Q j )
8: return P
This lets us adjust the number of tests via k. Each individual will be involved in at least k tests,
and our total number of tests will be k × || P||. This might be too much still. If we wanted to be a
little more pedantic, we could make sure that each individual had almost exactly k tests and the total
would be some value ≤ k × || P||. It needs some bookkeeping, and isn’t particularly random, but:
Algorithm 80 More Precise K-fold Relative Fitness Assessment

1: P ← population
3: R←P " Holds individuals who’ve not had enough tests yet
5: if Pi ∈ R then
6: R ← R − { Pi }
7: repeat
8: if || R|| < k then
9: Q ← R ∪ some || R|| − k unique individuals chosen at random from P − R, other than Pi
10: else
11: Q ← k unique individuals chosen at random from R
13: Test(Pi , Q j )
14: if Q j ∈ R and Q j has been involved in at least k tests then
15: R ← R − {Q j }
16: until Pi has been involved in at least k tests
19: return P
But this still might be too many. If fitness is done simply by counting number of games won
(as opposed to total score, etc.), another tack is to involve the entire population in a big single-
elimination tournament. The fitness of an individual is how high it rises in the tournament:
109
Algorithm 81 Single-Elimination Tournament Relative Fitness Assessment
1: P ← population " We presume || P|| is a power of 2
2: R ← P, shuffled randomly " To shuffle an array randomly, see Algorithm 26

3: for i from 1 to lg || P|| do " Oh, all right. lg is log2
4: Q←R " Q holds the current rank of undefeated individuals
5: R ← {} " R will hold the next rank of undefeated individuals
6: for j from 1 to || Q|| by 2 do
7: Test(Q j , Q j+1 )
8: if Q j defeated Q j+1 in that last Test then
9: R ← R ∪ {Q j }
10: else
11: R ← R ∪ { Q j +1 }
14: return P
This has exactly || P|| − 1 tests, and involves individuals in about 2 tests on average. But it has an
interesting and important feature that distinguishes it from Pairwise Relative Fitness Assessment
(Algorithm 77): individuals which are better are involved in more tests. In some sense this lets us be a
bit more discerning in distinguishing among our better players (who are more likely to be selected
anyway) than in distinguishing among the rabble. The disadvantage is that if the games are fairly
noisy, then a good player might be accidentally lost in the rabble. However, Single-Elimination
Tournament has often performed very well, if your desired test metric is simply based on games
won or lost (rather than points).86
Single-Elimination and Pairwise Relative Fitness Assessment bring up one cute additional
opportunity: fitnessless selection (a notion by Wojciech Jaśkowski, Krzysztof Krawiec, and Bartosz
Wieloch). Rather than use these methods to compute fitness, we might simply use them to determine
the winner in Tournament Selection without ever computing a fitness at all! For example, let’s
imagine we’re using Tournament Selection with a tournament of size 2. We need to select an
individual. To do so, we pick two individuals at random from the population and have them play a
game right then and there. The winner is the one selected. More specifically, if our tournament size
n is a power of 2, we could select n unique individuals at random from the population, and put
them through a little single-elimination tournament. The winner of the tournament is selected. We
could hash the tournament results to avoid replaying individuals in the off-chance they’ve played
against one another before.88
86 Yes, yes, you could construct a double-elimination tournament algorithm as well, or World-Cup soccer tournament
algorithm. But for heavens’ sakes, don’t model it after the BCS!87
87 http://en.wikipedia.org/wiki/Bowl Championship Series
88 In theory, this notion of lazy fitness assessment could be done for regular evolutionary computation too: only
assess the fitness of individuals once they’ve been entered into a Tournament Selection tournament. But the number of
individuals who are never picked, at random, for any tournament during the selection process, is probably pretty small,
so it’s not a huge advantage. Fitnessless Coevolution was proposed in Wojciech Jaśkowski, Krzysztof Krawiec, and
Bartosz Wieloch, 2008, Fitnessless coevolution, in Conor Ryan and Maarten Keijzer, editors, Genetic and Evolutionary
Computation Conference (GECCO), pages 355–362, ACM.
110
Sequential (or Serial) Parallel Parallel Previous
2-Population Competitive Coevolution 2-Population Competitive Coevolution 2-Population Competitive Coevolution
Generation Population P Population Q Population P Population Q Population P Population Q
4
13 13, 14
10 12 11 12 11 12
11 13 14
3
9 9, 10
6 8 7 8 7 8
7 9 10
2
5 5, 6
2 4 3 4 3 4
3 5 6
1
1 1, 2 1, 2
Fitness Assessment Population To Collaborating

Breeding is then
Be Assessed Population
Figure 41 Three different fitness assessment and breeding strategies for 2-Population Competitive Coevolution. Shown
are the two populations going through successive generations. Numbers indicate the order in which either breeding or
fitness assessment takes place.
Two more items of note. First, it may be possible to improve the gradient even further by playing
not against other individuals in the current generation, but against ones in, say, the previous
generation. See the discussion of Algorithm 87 for more on that trick. Second, 1-Population
Competitive Coevolution’s nice gradient-reduction feature is hardly a panacea. It’s still fairly easy
for the whole population to get stuck in notorious local optima. For example, if you’re looking for
an optimal soccer team, your initial population might contain terrible arbitrary players, but one or
two of them contain players which simply go to the ball and kick it into the goal (easy procedures).
These rack up such huge scores against the terrible players that soon the entire population consists
of teams of players who all go straight to the ball and try to kick it to the goal, and the population
just won’t budge from that local optimum, forcing you to rethink how to assess fitness.89
6.2 2-Population Competitive Coevolution

2-Population Competitive Coevolution finds robust solutions by simultaneously looking for good
solutions while searching for the places in the space of test cases that are most challenging. The idea
is simple and appealing: we construct not one but two populations (or if you like, subpopulations
or demes). Population P will contain the individuals we’re trying to robustify. Population Q will
contain test cases to challenge the individuals in P. The fitness of individuals in population P
will be based on how well they perform against the individuals in population Q, and likewise,
the individuals in Q will be assessed based on how well they perform against the individuals in
population P.
I call P the primary population and Q the alternative or foil population. Ultimately we’re re-
ally just interested in the primary population. Thus, unlike 1-Population Competitive Coevolution,
we only bother to assess the external fitness of population P, and likewise only maintain the best
individuals from P. Rarely are the best of Q interesting. When we test one population in the context
of the other (be it P or Q), the other population is known as the collaborating population.
89 This happened to me.
111
The classic illustration of 2-Population Competitive Coevolution is Danny Hillis’s90 attempt to
discover optimal sorting networks. A sorting network is a series of comparisons on elements in
the array which, when completed, results in the array being sorted. Each comparison compares
two fixed elements in the array. If the upper item is smaller than the lower item, then the two items
are swapped. Some comparisons can be done in parallel, and the whole mechanism is attractive
because it can be done in hardware. The objective is not only to find a correct sorting network, but
one which has the fewest comparisons possible.
The problem of finding an optimal sorting network is one
which requires test cases: there are lots and lots of different ar-
rays of numbers to test against, and you need to find a network
which works properly with all of them. Hillis began with a
population of sorting networks and tested each one against a
random sample of test cases. However, a random sample was
unlikely to provide those pathological test cases that would be Figure 42 A sorting network for an ar-
particularly difficult for an individual. To deal with this, Hillis ray of four numbers. Vertical lines indi-
cate comparisons. Progresses left to right.
turned to 2-Population Competitive Coevolution: he coevolved Two of the comparisons can be done in
a population of sorting networks against a competing popu- parallel.
lation of hard-to-sort arrays of numbers. The fitness of a sorting
network was the number of arrays it got correct; and the fitness of an array was the number of
sorting networks it stumped. Thus while the sorting networks were improving themselves, the
arrays were finding harder corner cases to challenge the sorting networks.
Because it has mutliple populations, 2-Population Competitive coevolution introduces new
ways to interleave fitness assessment and breeding. Three are shown in Figure 41. The first way
is to assess the internal fitness of P in the context of Q (and also assess P’s external fitness), then
breed P, then assess the internal fitness of Q in the context of the new P, then breed Q, and so forth.
This is known as a sequential or serial 2-Population Competitive Coevolutionary Algorithm:
Algorithm 82 An Abstract Sequential 2-Population Competitive Coevolutionary Algorithm
2: Q ← Build Initial Alternative Population
3: Best ← !
4: repeat
5: AssessInternalFitness(P, Q)
6: AssessExternalFitness(P) " Used to determine algorithm progress and Best.
9: Best ← Pi
11: AssessInternalFitness(Q, P)
12: Q ← Join(Q, Breed(Q))
14: return Best
90 Google his name. Danny Hillis invented the coolest supercomputer ever. Hillis’s coevolution paper is Daniel Hillis,
1990, Co-evolving parasites improve simulated evolution as an optimization procedure, Physica D, 42(1–3). It’s a famous
paper, but it has some oddities, including a pretty unusual diploid representation with two copies of each gene.
112
This requires some way to assess the internal fitness of a population in the context of its
collaborating population. The straightforward way to do it is to sample k individuals from the
collaborating population to test against:
Algorithm 83 K-fold Relative Fitness Assessment with an Alternative Population

1: M ← population to be Tested
2: C ← collaborating Population " Either one could be P or Q, depending

5: W ← k unique individuals chosen at random from C
6: for each individual Wj ∈ W do
7: Test(Mi , Wj )
8: AssessFitness(Mi ) " Using the results of all Tests involving Mi
9: return M
Notice that we didn’t call the two populations P and Q, but rather M and C, because either one
could be P (and the other Q), depending on which population is being assessed.
Sequential 2-Population Competitive Coevolution has two downsides which make it less than
attractive. First, Q must be tested against the new, improved P, always one step ahead of it. Second,
the assessments are separated, which means you can’t combine AssessInternalFitness(P, Q) and
AssessInternalFitness(Q, P). Probably when you were testing P, you got some fitness information
about individuals in Q as well. Why throw that information away? This leads to our second
method: a parallel 2-Population Competitive Coevolutionary Algorithm. Here, each population is
tested against the other, and then both of them breed. This solves both of these problems: neither
population has a leg up on the other by design, and we can group internal fitness assessment
together for both populations:
Algorithm 84 An Abstract Parallel 2-Population Competitive Coevolutionary Algorithm

3: Best ← !
4: repeat
5: AssessInternalFitness(P, Q) " Internal fitness assessment could be done simultaneously
7: AssessExternalFitness(P) " Used to determine algorithm progress and Best.
10: Best ← Pi
12: Q ← Join(Q, Breed(Q))
14: return Best
113
We could do the internal fitness tests separately still if we liked. But if we wished, we could test
them together. For example, if the two populations were the same size, we could shuffle P, then test
each individual Pi against the corresponding individual Qi . To do further tests, we might test each
Pi against Qi+1 , then Qi+2 , and so on, wrapping around as necessary. But this creates statistical
dependencies among the tests: for example, individuals Pi and Pi+1 would be tested against almost
exactly the same individuals, which is probably not good. Instead we could shuffle the population
P each time, but then we’d like to guarantee that in the shuffling certain individuals never get
paired up again if they’ve been tested together before. A quick-and-dirty, not particularly elegant
solution to that is to shuffle P each time, then as long as there is a pair that’s already been tested
before, we break up that pair. It’s a hack. Here we go:
Algorithm 85 K-fold Relative Joint Fitness Assessment with an Alternative Population

1: P ← population
2: Q ← alternative Population
4: P) ← P, shuffled randomly " To shuffle an array randomly, see Algorithm 26

5: for j from 1 to k do
6: for i from 1 to || P|| do " We assume that || P|| = || Q||
7: Test(Pi) , Qi )
8: Shuffle P) randomly
9: while there is a value i where Pi) and Qi have already been tested together do
10: Swap Pi) with some randomly chosen individual Pl)
11: for i from 1 to || P|| do
12: AssessFitness(Pi) ) " Using the results of all Tests involving Pi)
13: AssessFitness(Qi ) " Using the results of all Tests involving Qi
14: return P, Q
External fitness assessment can be a problem for both of these options (if you care about doing
it). You can’t test against Q per se, because Q keeps changing (and ideally improving), and so
you won’t get a consistent, absolute fitness metric for P. It’s the same conundrum that occurs in
1-Population Competitive Coevolution. Perhaps you could create a fixed sample drawn from the
test-case space and test against that; or create a guru of some sort.
One last option is to assess each population against the previous generation of the collaborating
population. This might help improve the gradient a bit because each population is given a bit easier
time. Except for the first generation, we’re back to testing populations separately again. I call this
the Parallel Previous 2-Population Competitive Coevolutionary Algorithm.91
91 Parallel Previous brings up one interesting approach to doing external fitness assessment: report how well you
defeated the previous generation. This fitness metric is essentially measuring the slope of your fitness improvement: if
it’s positive, you’re making progress.
114
Algorithm 86 An Abstract Parallel Previous 2-Population Competitive Coevolutionary Algorithm
3: Best ← !
4: AssessInternalFitness(P, Q) " Internal fitness assessment could be done simultaneously
6: AssessExternalFitness(P)
9: Best ← Pi
10: repeat
11: P) ← Join(P, Breed(P)) " We do this to let us test against the previous generation
12: Q) ← Join(Q, Breed(Q)) " Ditto
13: )
AssessInternalFitness(P , Q)
14: AssessInternalFitness(Q) , P)
15: AssessExternalFitness(P) )
17: if ExternalFitness(Pi ) > ExternalFitness(Best) then
18: Best ← Pi
19: P ← P)
20: Q ← Q)
22: return Best
There is one nifty thing you can do with the Parallel Previous version: because individuals are
being tested against last generations’ individuals, we know those previous generations’ individuals’
fitnesses already. That means that we could choose to test not just against random individuals but
against, say, the fittest individuals of the previous population. To test against the fittest individuals
of a collaborating population:
Algorithm 87 K-fold Relative Fitness Assessment with the Fittest of an Alternative Population
1: M ← population to be Tested " Either P or Q could be M or C
2: C ← collaborating Population " Individuals in C already have their fitnesses assessed
4: C ) ← C sorted by fitness, fittest individuals first

6: for j from 1 to k do
7: Test(Mi , Cj) )
8: AssessFitness(Mi ) " Using the results of all Tests involving Mi
9: return M
Of course, you could do a mix of the Parallel and Parallel Previous methods: test an individual
against some k individuals from the current-generation collaborating population, and also test the
115
individual against the n fittest individuals from the previous-generation collaborating population.
This could also work with 1-Population Competitive Coevolution as well. In that case, M is the
current generation of the individuals and C is the previous generation of the same individuals.
Arms Races and Loss of Gradient 2-Population Competitive Coevolution is often viewed as
an abstract version of a biological arms race: one population learns a trick, forcing the second
population to learn a new trick to beat the first one, and so on. In an ideal world, the arms race
results in a natural, gradual build-up of gradient, so we’re not faced with the Needle-in-a-Haystack
scenario as had been shown in Figure 40.
Sadly, this is often not the case. Instead, one popu- Fitness
lation may have an easier optimization task, and so it
improves so rapidly that it leaves the other population The Better Population
in the dust. At some point all the individuals in one
population (say, Q) are so good that they all defeat the
individuals in P soundly. When this happens, all the
individuals in Q now basically have all the same fitness,
because they all beat everyone in P. Likewise all the
individuals in P have the same fitness because they all
lose to everyone in Q. The Worse Population
This condition is called loss of gradient: the selec-

Time
tion operator no longer has anything to go on, and starts
picking individuals at random. This usually causes the Figure 43 Loss of Gradient. One population peri-
external fitness to start dropping until the populations odically improves so much that selection starts to
fail, causing a drop in fitness.
reestablish gradient again, resulting in the maddening
situation shown in Figure 43. This isn’t an easy thing
to fix, though here’s one strategy: if you can somehow detect that a population is improving too
rapidly and loss of gradient is seeping in, you might pause the evolution of that population until
gradient is reestablished. Of course, you’d need to think about how to detect that: perhaps lower
variance among the fitnesses in your populations?
You might also try using the Parallel Previous variation: I suspect (but have no evidence that) it
will dampen loss of gradient a bit.
6.3 N-Population Cooperative Coevolution

Competitive coevolution tries to improve individuals by pitting them against other individuals.
In contrast, cooperative coevolution, proposed by Mitchell Potter and Ken De Jong,92 strives to
find individuals who work well together. Why would you want to do that? The issue is this: some
optimization spaces are high-dimensional and gigantic. Cooperative coevolution simplifies those
spaces by breaking them into multiple, much simpler, subspaces for different populations to search.
92 I think the earliest of their publications on the matter is Mitchell Potter and Kenneth De Jong, 1994, A cooperative
coevolutionary approach to function optimization, in Yuval Davidor, Hans-Paul Schwefel, and Reinhard Manner, editors,
Proceedings of the Third Conference on Parallel Problem Solving from Nature, pages 249–257. The two then more fully fleshed it
out in Mitchell A. Potter and Kenneth A. De Jong, 2000, Cooperative coevolution: An architecture for evolving coadapted
subcomponents, Evolutionary Computation, 8(1), 1–29.
116
Let’s say you’re looking for an optimal robot soccer team. Each team has eleven different
players, all of whom you believe must have unique robot behaviors93 There are 1000 different
behaviors for each robot. As a result, you are trying to find an optimal setting in a space of 100011
(or 1000000000000000000000000000000000) possibilities. That’s a big number.
Here’s what you could do instead. For each robot, create an arbitrary set of behaviors for
everyone but that robot. Then use an optimization method to find the optimal robot behavior given
his arbitrary team of fixed collaborating robots. At the end, take the optimal robots and put them
together to form a final team. This requires 11 optimization runs, each of which is over a simple
space of only 1000 possibilities. However there’s no guarantee that those robots will work together
well at the end: after all they were not designed for one another but rather were designed to be the
best robot given the arbitrary team they were forced to work with.
N-Population Cooperative Coevolution strikes a middle-ground between these two situations.
We perform 11 optimization runs at one time, but individuals in each run are tested by grouping
them with the current individuals from the other runs. We do this by creating 11 populations: a
population of goalies, a population of left fullbacks, a population of right fullbacks, ... etc. Each
population has its own optimization process. When an individual from a population needs to
be tested as part of his fitness assessment, we group it with individuals selected from the other
populations to form a complete soccer team. We then play a match, and the match results become
the individual’s test results.
In short, each of the populations in N-Population Cooperative Coevolution is finding an optimal
subsolution: a sub-part of the complete solution. This allows us to break a search space of size n a
into n search spaces each of size a, a huge reduction in complexity. In 2-Population Cooperative
Coevolution, there was one specific population of interest to us. But now all the populations are
of interest to us because they are each solving a piece of the problem. We can easily construct an
external fitness to gauge the system: it’s the fittest collaboration we’ve found so far among the
individuals in the various populations.
N-Population Cooperative Coevolution can be done in any of the same frameworks that were
shown in Figure 41 for 2-Population Competitive Coevolution, but in truth, we only really see the
Sequential and Parallel methods in practice. Whereas the Sequential method has problems for
Competitive Coevolution, it’s a fine technique for Cooperative Coevolution, assuming that we have
the time to waste on additional tests.94 So rather than show Internal and External fitness assessment,
we’ll just assess a joint fitness, and then store in Best the fittest joint vector of individuals, one from
each population, that we’ve found so far.
93 If you were smart about it, you might instead just have four behaviors: one for the goalie, one that all defenders use,
one that all midfielders uses, and one that all attackers use. That’d be a lot simpler space to search.
94 Sequential N-Population Cooperative Coevolution is an example of an Alternating Optimization (AO) algorithm.
AO algorithms presume that you are trying to optimize a function f (!x ) by breaking !x into various smaller variables
$ x1 , ..., xn %. To start, you assign arbitrary values to each of the xi . Then, you optimize x1 while holding the other xi
fixed. Once x1 is sufficiently optimized, you fix it to its new value, and now optimize x2 while holding all other xi
fixed. Continue this process up through xn . Then repeat again, optimizing x1 with the others fixed, and so on. Sound
familiar? AO doesn’t presume any particular optimization method for each of the xi : it’s just a framework for a variety
of algorithms. If you’re interested, other (famous but non-metaheuristic) algorithms which fall squarely under the AO
banner are Expectation Maximization (EM) and its degenerate variation k-means clustering, techniques for finding
clusters in data.
117
Algorithm 88 An Abstract Sequential N-Population Cooperative Coevolutionary Algorithm (CCEA)
1: P(1) , ..., P(n) ← Build n Initial Populations
−→
2: Best ← !
3: repeat
5: AssessJointFitness($i %, P(1) , ..., P(n) ) " Computes fitness values for only population P(i)
(1) (n) (1)
6: for each vector !s of individuals $ Pa , ..., Pz %: Pa ∈ P(1) , etc., assessed in Line 5 do
−→ −→
7: if Best = ! or JointFitness(!s) > JointFitness(Best) then
−→
8: Best ← !s
9: P(i) ← Join(P(i) , Breed(P(i) ))
−→
−→
11: return Best
Note that in the For-loop we assess some joint fitnesses but only apply them to the individuals
in population P(i) . We could do that with a variant of algorithm 83 which works like this. For
each individual in P(i) we perform some k tests by grouping that individual with randomly-chosen
individuals from the other populations to form a complete solution:
Algorithm 89 K-fold Joint Fitness Assessment with N − 1 Collaborating Populations

1: P(1) , ..., P(n) ← populations
2: i ← index number of the Population to be Tested
4: !s ← $s1 , ..., sn % an (empty for now) complete solution " We’ll fill it up with individuals
(i ) i
5: for each individual Pj ∈ P do ( ) " For each individual to test...
6: for w from 1 to k do " Do k tests...
7: for l from 1 to n do " Build a complete solution including the individual to test
8: if l = i then " It’s the individual to test
(l )
9: sl = Pj
10: else " Pick a random collaborator
11: sl = individual chosen at random from P ( l )
12: Test(!s) " Test the complete solution

(i ) (i )
13: AssessFitness(Pj ) " Using the results of all Tests involving Pj
14: return P(1) , ..., P(n)
We’ve abandoned here any attempt of using unique collaborators: but you can do that if you
really want to try it. I don’t think it’s that valuable because the space is so much larger. The
Sequential approach is the original method proposed by Potter and De Jong, and it still remains
popular. But, in the formulation described above, it’s wasteful because we do many tests but only
use them to assess the fitness of a single individual — the collaborators are forgotten about. We
could fix that by keeping around the previous tests and including them when we get around to
testing the collaborating individuals for their fitness assessment. Or we could just do the Parallel
approach. Specifically, we test everyone together, then breed everyone at once:
118
Algorithm 90 An Abstract Parallel N-Population Cooperative Coevolutionary Algorithm
1: P(1) , ..., P(n) ← Build n Initial Populations
−→
2: Best ← !
3: repeat
4: AssessJointFitness($1, ..., n%, P(1) , ..., P(n) ) " Computes fitness values for all populations
(1) (n) (1)
5: for each vector !s of individuals $ Pa , ..., Pz %: Pa ∈ P(1) , etc., assessed in Line 4 do
−→ −→
6: if Best = ! or JointFitness(!s) > JointFitness(Best) then
−→
7: Best ← !s
9: P(i) ← Join(P(i) , Breed(P(i) ))
−→
−→
11: return Best
This doesn’t look like a big change, but it is. Because we can group all the joint fitnesses together
at one time, we can save some testing time by not doing further tests on collaborators who’ve been
involved in a sufficient number of tests already. We could do this with a variation of Algorithm 85,
but with N > 2 it might suffice to just pick collaborators at random, even if some by chance get
tested more than others, hence:
Algorithm 91 K-fold Joint Fitness Assessment of N Populations

1: P(1) , ..., P(n) ← populations
3: !s ← $s1 , ..., sn % an (empty for now) complete solution " We’ll fill it up with individuals
4: for i from 1 to n do " For each population...
(i )
5: for each individual Pj ∈ P(i) do " For each individual in that population...
(i )
6: m ← number of tests individual Pj has been involved in so far
7: for w from m+1 to k do " Do at most k tests...
8: for l from 1 to n do " Build a complete solution including the individual to test
9: if l = i then " It’s the individual to test
(l )
10: sl = Pj
11: else " Pick a random collaborator
12: sl = individual chosen at random from P(l )
13: Test(!s) " Test the complete solution
(i )
15: for each individual Pj ∈ P(i) do
(i ) (i )
16: AssessFitness(Pj ) " Using the results of all Tests involving Pj
17: return P(1) ...P(n)
119
Pathological Conditions in Testing So what could go wrong? For one, there’s the theoretical
possibility of laziness. If certain populations are doing impressively, other populations may just
come along for the ride. For example, let’s say you’re trying to find an optimal team of basketball
players. You’ve got a population of centers, of forwards, of guards, etc. Your guard population has
converged largely to consist of copies of Michael Jordan. The Michael Jordans are so impressive that
the population of (say) forwards doesn’t need to do any work for the team to be near optimal. In
essence, all the forwards’ fitnesses look the same to the system: regardless of the forward selected,
the team does really really well. So the system winds up selecting forwards at random and the
forwards don’t improve. This condition is the cooperative equivalent of the Loss of Gradient
pathology discussed earlier. The basic solution to this is to change your fitness function to be
more sensitive to how the forwards are doing. For example, you might apply some kind of credit
assignment scheme to assign the fitness differently to different cooperating individuals. Be careful:
the system is now likely no longer cooperative, that is, coordinating individuals no longer receive
the same fitness, and this can result in unexpected dynamics.
Laziness is the tip of the iceberg though. How do
you assess the fitness of a cooperative coevolutionary
Individual A1
individual based on tests? Early on it was thought that
Broad
you might base it on the average of the test results with Suboptimum
Population A
various collaborators from the other population(s). Let’s
Joint Space
say that there is one optimal joint solution, but the hill
leading to it is very small; whereas there’s a large sub- Individual A2
optimal peak elsewhere, as in Figure 44. If we tested
individuals A1 and A2 with many individuals from Pop- Narrow Optimum
ulation B and took the average, A1 would appear fitter

on average even though A2 was actually a collaborator
Population B
in the optimum. A1 is a jack-of-all-trades-but-master-of-
none individual which is never phenomenal anywhere, Figure 44 Relative Overgeneralization.
but most of the time it’s involved in a joint solution
that’s better than average.
This situation leads to a pathological condition called relative overgeneralization, where the
populations converge to joint solutions which are suboptimal, but involve lots of jacks-of-all-trades.
Paul Wiegand discovered this unfortunate situation.95 The way to fix this is to assess fitness as the
maximum of the tests rather than their average. However to get good results you may need to do a
lot of tests, perhaps even against the entire other population. It turns out that usually there are just
a few “special” collaborators in the other population(s) which, if you tested just with them, would
compute fitness orderings for your entire population in exactly the same way as testing against
everyone. Liviu Panait, a former student of mine, developed a 2-population cooperative algorithm,
iCCEA, which computes this archive of special collaborators, resulting in far fewer tests.96
Finally, if your fitness function has multiple global optima, or near-optima, you could also
wind up victim to miscoordination.97 Let’s say you have two cooperating populations, A and B,
95 See Paul’s thesis: R. Paul Wiegand, 2004, An Analysis of Cooperative Coevolutionary Algorithms, Ph.D. thesis, George
Mason University, Fairfax, Virginia.

96 See his thesis: Liviu Panait, 2006, The Analysis and Design of Concurrent Learning Algorithms for Cooperative Multiagent
Systems, Ph.D. thesis, George Mason University, Fairfax, Virginia.

97 Miscoordination isn’t a disaster: an explorative enough system will find its way out. But it’s worthwhile mentioning
that it is a disaster in a sister technique in artificial intelligence, multiagent reinforcement learning.
120
and two global optima, 1 and 2. The two optima are off-
set from one another as shown in Figure 45. Population
A has discovered an individual A1 who is part of global
Individual A1
optimum 1 (yay!), and likewise Population B has discov-
ered an individual B2 who is part of global optimum 2 Suboptimal
Population A
Global Optimum 1 Region
(yay!). But neither of these individuals will survive, be-
cause Population A hasn’t yet discovered individual A2 Joint Space
Individual A2
who, when collaborating with B2, would help B2 shine.
Likewise Population B hasn’t yet found individual B1
who would make A1 look great. In the worst case, these Global Optimum 2
populations are trying out A1 and B2 in combination,
which winds up in a quite suboptimal region of the joint
space. Thus, though A1 and B2 are optimal for their Population B
respective populations, the populations can’t tell: they Individual B2

Individual B1
look bad.
Figure 45 Miscoordination.
6.4 Niching: Diversity Maintenance Methods
To add exploration in your system, perhaps to prevent it from converging too rapidly to suboptimal
solutions, there are many options available. So far we’ve considered:
• Increasing your sample (population) size
• Adding noise to your Tweak procedure
• Being less selective among individuals (picking less fit ones more often)
• Adding random restarts to your system
• Adding explicit separation constraints in your population (as is done in various parallel
stochastic optimization approaches like Island Models or Spatially-embedded Models)
• Explicitly trying to add different individuals from the current ones in the population (as is
done in Scatter Search with Path Relinking)
One approach we’ve not yet considered is to punish individuals in some way for being too
similar to one another. For example, we might explicitly lower the fitness of individuals if they’re
too close to other individuals (fitness sharing). Or we could pick individuals to die based on how
similar they are to new incoming children in a steady-state or generation-gap algorithm (crowding).
These approaches all affect the survivability of individual A (versus individual B) based on whether
or not there exists some individual C (which is similar to A), in the population already, or being
introduced new to the population. Thus these methods are coevolutionary in nature.98
Before we examine techniques, we need to consider what similar means. Two individuals can
be similar in at least three ways:
98 One additional diversity maintenance approach we won’t really discuss here — it’s not coevolutionary in nature — is
incest prevention. Here, individuals are not permitted to cross over with other individuals if they share a parent (or a
grandparent, or however deep you’d like to go). There has also been a bit of work on what I call explicit speciation,
where each individual has a small tag which indicates its “species” (the tag can be mutated), and selection or breeding is
constrained in some way to be mostly within species. This usually is for other purposes than diversity maintenance.
121
• Phenotypically: they behave similarly.
• Genotypically: they have roughly the same makeup when it comes to breeding.
• Individuals may have similar fitness.
Ideally we’re looking for a phenotypical similarity: but often it’s not easy to determine what
that is exactly — or perhaps your phenotypes and genotypes are basically identical. So often one
settles on some notion of genotypical similarity. Fitness similarity makes no sense in this context:
but when we get to multi-objective algorithms (which have more than one fitness measure), it will
suddenly make lots of sense!
To determine how similar individuals are, we’ll need some kind of distance measure which
ideally defines a metric distance99 in the phenotypical (or genotypical) space. If your individuals
already reside in a metric space, you’re in luck. For example, if your individuals are vectors of
real-valued numbers (individual i has the genotype $i1 , ..., in % and individual j has the genotype
$ j1 , ..., jn %), and you’re making the assumption that genotype distance is the same ) as phenotype
distance, then you might use the sum squared genotype distance, that is, d(i, j) = ∑k (ik − jk )2 .
For boolean vectors, you could use the Hamming distance, which counts the number of times that
two genes are different, that is, d(i, j) = ∑k ik ⊕ jk , where ⊕ is the XOR (exclusive OR) operator. If
your individuals are more complex — trees, say — have a lot of fun defining a distance measure
among them!
6.4.1 Fitness Sharing

The idea behind fitness sharing is to encourage diversity in individuals by reducing their fitness
for being too similar to one another.100 The most common form of fitness sharing, proposed by
David Goldberg and Jon Richardson, requires you to define a neighborhood radius σ. We punish a
given individual’s fitness if there are other individuals within that radius. The more individuals
inside that radius, and the closer the individuals are to the given individual, the worse its fitness.
Given our distance function d(i, j), we compute sharing function s between two individuals i
and j, which tells us how much punishment i will receive for j being near it:
(
1 − (d(i, j)/σ)α if d(i, j) < σ
s(i, j) =
0 otherwise
α > 0 is a tuning parameter you can set to change the degree of punishment i receives for j
being particularly close by. The size of σ is tricky: too small and the force for diversity is weak; but
99 A metric space is a space where we can construct a distance measure which obeys the triangle inequality. More
specifically, the distance function d(i, j) must have the following properties. First, it should always be ≥ 0 (what’s
a negative distance?). Second, it should be 0 only if i = j. Third, the distance from i to j should be the same as the
distance from j to i. And last, the triangle inequality: for any three points i, j, and k, it must always be true that
d(i, k) ≤ d(i, j) + d( j, k). That is, going from point i to point k directly is always at least as short as taking a detour
through j. Metric spaces include ordinary multi-dimensional real-valued Euclidian space and the space of boolean
vectors (using Hamming distance). But what’s the metric space of trees? Does one even exist?
100 The term “fitness sharing” is unfortunate: they’re not sharing fitness with one another. They’re all just having their
fitnesses reduced because they’re too close to one another. The technique was first discussed, I believe, in David Goldberg
and Jon Richardson, 1987, Genetic algorithms with sharing for multimodal function optimization, in John J. Grefenstette,
editor, Proceedings of the Second International Conference on Genetic Algorithms, pages 41–49, Lawrence Erlbaum Associates.
122
it shouldn’t be so large that multiple optima fall in the same neighborhood (or even close to that).
Now we adjust the fitness as follows:
(ri ) β
fi =
∑ j s(i, j)
ri is the actual (raw) fitness of individual i and f i is the adjusted fitness we will use for the
individual instead. β > 1 is a scaling factor which you’ll need to tune carefully. If it’s too small,
individuals won’t move towards optima out of fear of crowding too near one another. If it’s too
large, crowding will have little effect. Of course you probably don’t know much about the locations
of your optima (which is why you’re using an optimization algorithm!), hence the problem. So
there you have it, three parameters to fiddle with: α, β, and σ.
If your fitness assessment is based on testing an individual against a bank of test problems (for
example, seeing which of 300 test problems it’s able to solve), you have another, simpler way to
do all this. Robert Smith, Stephanie Forrest, and Alan Perelson have proposed an implicit fitness
sharing:101 if an individual can perform well on a certain test case and few other individuals can
do so, then the individual gets a big boost in fitness. The approach Smith, Forrest, and Perelson
took was to repeatedly sample from the population over and over again, and base fitness on those
samples. In Implicit Fitness Sharing, you must divide the spoils with everyone else who did as
well as you did on a given test.
Algorithm 92 Implicit Fitness Sharing
1: P ← population
2: k ← number of times we should sample " Should be much bigger than || P||
3: σ ← how many individuals per sample
4: T ← test problems used to assess fitness
5: C ← || P|| by || T || matrix, initially all zeros " Ci,j is how often individual Pi was in a sample for Tj
6: R ← || P|| by || T || matrix, initially all zeros " Ri,j is individual Ri ’s sum total reward for Tj
7: for each Tj ∈ T do
8: for k times do
9: Q ← σ unique individuals chosen at random from P
10: for each individual Ql ∈ Q do
11: i ← index of Ql in P
12: Ci,j ← Ci,j + 1
13: S ← individual(s) in Q which performed best on Tj " Everyone in S performed the same
14: for each individual Sl ∈ S do
15: i ← index of Sl in P
16: Ri,j ← Ri,j + 1/||S||
17: for each individual Pi in P do
18: Fitness(Pi ) ← ∑ j Ri,j /Ci,j
19: return P
101 Thiswas part of a larger effort to develop optimization algorithms fashioned as artificial immune systems. The
authors first suggested it in Robert Smith, Stephanie Forrest, and Alan Perelson, 1992, Population diversity in an
immune system model: Implications for genetic search, in L. Darrell Whitley, editor, Proceedings of the Second Workshop on
Foundations of Genetic Algorithms, pages 153–165, Morgan Kaufmann.
123
Note that it’s possible that an individual will never get tested with this algorithm, especially if k
is too small: you will want to check for this and include the individual in a few tests.
Believe it or not, this is quite similar to fitness sharing: the “neighborhood” of an individual is
phenotypical: those individuals who solved similar test problems. You’ll again need a “neighbor-
hood radius” σ. But this time instead of defining an explicit radius in phenotype space, the “radius”
is a sample size of individuals that compete for a given test problem t. You’ll need to fiddle with
the new σ as well, but it’s likely not as sensitive. k is a parameter which should be as large as you
can afford (time-wise) to get a good sample.
6.4.2 Crowding
Crowding doesn’t reduce the fitness of individuals for being too similar; rather it makes them more
likely to be picked for death in a steady-state system. Though steady-state evolution is usually
exploitative, the diversity mechanism of crowding counters at least some of that. The original
version of crowding, by Ken De Jong,102 was similar to a steady-state mechanism: each generation
we breed some n new individuals. Then one by one we insert the individuals in the population,
replacing some individual already there. The individual selected to die is chosen using Tournament
Selection not based on fitness but on similarity with the individual to insert. Note that because of the
one-by-one insertion, some of the individuals chosen to die might be some of those n children; so
this isn’t quite a steady-state algorithm. But it’s fine to do crowding by using a plain-old steady-state
algorithm with selection for death based on similarity to the inserted child.
As it turns out, crowding doesn’t perform all that well. But we can augment it further by
requiring that the child only replaces the individual chosen to die if the child is fitter than that
individual. This approach is called Restricted Tournament Selection,103 by Georges Harik, and
seems to work pretty well.
Samir Mahfoud proposed an entirely different mechanism, Deterministic Crowding,104 in
which we randomly pair off parents in the population, then each pair produces two children.
Each child is matched with the parent to which it is most similar. If the child is fitter than its
matched parent, it replaces the parent in the population. The idea here is to push children to replace
individuals (in this case, their own parents) which are similar to them and aren’t as fit as they are.
Mahfoud’s formulation is an entire generational evolutionary algorithm instead of simply a fitness
assessment mechanism:
102 From his thesis, Kenneth De Jong, 1975, An Analysis of the Behaviour of a Class of Genetic Adaptive Systems, Ph.D. thesis,
University of Michigan. The thesis is available online at http://cs.gmu.edu/∼eclab/kdj thesis.html
103 Georges Harik, 1995, Finding multimodal solutions using restricted tournament selection, in Larry J. Eshelman,
editor, Proceedings of the 6th International Conference on Genetic Algorithms, pages 24–31, Morgan Kaufmann.
104 Mahfoud first mentioned this in Samir Mahfoud, 1992, Crowding and preselection revisited, in Reinhard Männer
and Bernard Manderick, editors, Parallel Problem Solving From Nature II, pages 27–36, North-Holland. But it actually got
fleshed out in his thesis, Samir Mahfoud, 1995, Niching Methods for Genetic Algorithms, Ph.D. thesis, University of Illinois
and Urbana-Champaign.
124
Algorithm 93 Deterministic Crowding
2: P ← {}
5: Best ← !
9: Best ← Pi
10: repeat
11: Shuffle P randomly " To shuffle an array randomly, see Algorithm 26
12: for i from 1 to || P|| by 2 do
13: Children Ca , Cb ← Crossover(Copy(Pi ), Copy(Pi+1 ))
16: AssessFitness(Ca )
19: Best ← Ca
21: Best ← Cb
22: if d(Ca , Pi ) + d(Cb , Pi+1 ) > d(Ca , Pi+1 ) + d(Cb , Pi ) then
23: Swap Ca and Cb " Determine which child should compete with which parent
24: if Fitness(Ca ) > Fitness(Pi ) then " Replace the parent if the child is better
25: Pi ← Ca
26: if Fitness(Cb ) > Fitness(Pi+1 ) then " Replace the parent if the child is better
27: Pi+1 ← Cb
29: return Best
125
126
7 Multiobjective Optimization
It’s often the case that we’re not interested in optimizing a single fitness or quality function, but
rather multiple functions. For example, imagine that a building engineer wants to come up with
an optimal building. He wants to find buildings that are cheap, tall, resistant to earthquakes, and
energy efficient. Wouldn’t that be a great building? Unfortunately, it might not exist.
Each of these functions to optimize is known as an
objective. Sometimes you can find solutions which are Dominated by A
Cheaper
optimal for every objective. But more often than not, ob-
jectives are at odds with one another. Your solutions are
Solution A
thus often trade-offs of various objectives. The building en-
gineer knows he can’t find the perfect building: cheap, tall, Dominated by A
strong, green. Rather, he might be interested in all the best
options he has available. There are lots of ways of defining
a set of “best options”, but there’s one predominant way:
the Pareto105 front of your space of candidate solutions.
Let’s say you have two candidate buildings, M and N. More Energy Efficient
M is said to Pareto dominate N if M is at least as good as
Figure 46 Region of solutions Pareto domi-
N in all objectives, and superior to N in at least one objec- nated by solution A, including the solution on
tive. If this were the case, why would you ever pick N the border. Keep in mind that this is not a de-
instead of M? M is at least as good everywhere and bet- piction of the phenotype space, but rather results
ter in something. If we have just two objectives (Cheaper, for the two objectives.
More Energy Efficient) Figure 46 shows the region of space
dominated by a given building solution A. The region is
“nearly closed”: the border is also dominated by A, except
Cheaper
Pareto
the corner (individuals identical to A in all objectives). Nondominated
Front
Neither M nor N dominates the other if they’re iden-
tical in all objectives, or if N is better in some things but
M is better in other things. In those cases, both M and N
are of interest to our building engineer. So another way of
saying the “best options” is the set of buildings which are
dominated by nobody else. We say that these buildings are
nondominated. This set of buildings is the Pareto front of More Energy Efficient
the space of solutions. Figure 47 at right shows the Pareto
Figure 47 The Pareto front of nondominated
front of the possible solutions in our two-objective space.
solutions.
Pareto fronts define outer borders. In a two-objective situa-
tion the Pareto front is often a curve demarcating that outer border. In a three-objective situation
it’s a skin of sorts. If you have one solution which is clearly superior to all the others (a superman,
so to speak), the front collapses to that single individual.
As shown in Figure 48, Pareto fronts come in different flavors. Convex fronts are curved
outwards towards better solutions. Concave fronts are curved inwards away from better solutions.
Nonconvex fronts aren’t entirely convex, and they include concave fronts as a subcategory. Fronts
105 Vilfredo Pareto (1848–1923) was an Italian economist responsible for a lot of important economics mathematics
concepts, including Pareto’s Law of income distribution, the 80–20 Rule (80% of events happen from only 20% of causes,
so you can fix most of your problems by focusing on just a few issues), and Pareto Efficiency and Pareto Optimality,
which is what we’re discussing here.
127
can also be discontinuous, meaning that there are regions along the front which are simply impos-
sible for individuals to achieve: they’d be dominated by another solution elsewhere in the valid
region of the front. There also exist locally Pareto-optimal fronts in the space where a given point,
not on the global Pareto front, happens to be pareto-optimal to everyone near the point.
Spread It’s not enough to offer our building engineer 100
Cheaper
points that lie on the Pareto front. What if they’re all in one
far corner of the front? That doesn’t tell him much at all Convex
Concave
about the options he has available. More likely he wants (Also Non-Convex)
samples that are spread evenly across the entire front. Thus Non-Convex
many of the algorithms that optimize for Pareto fronts also Discontinuous
try to force diversity measures. But interestingly, the dis-
tance measures used are rarely with regard to genotypical
or phenotypical distance; rather they’re distance in fitness:
how far are the candidate solutions away from each other More Energy Efficient
in the multi-objective space? This turns out to be much sim- Figure 48 Four kinds of Pareto fronts.
pler to compute than genotypical or phenotypical distance.
The Problem of Too Many Objectives As the number of objectives grows, the necessary size of
the populations needed to accurately sample the Pareto front grows exponentially. All the methods
in this section face certain challenges when scaling to large numbers of objectives (and by “large”
I mean “perhaps more than 4”). It’s a difficulty stemming from the nature of the problem itself.
To counter this, researchers have lately been turning to more exotic techniques, particularly ones
centering around the hypervolume covered by the Pareto front; but these techniques are both
complex and generally of high computational cost. We’ll focus on the more basic methods here.
A Note on Defining Fitness It is traditional in multiobjective optimization literature to define

fitness in terms of error. That is, the lower the objective value, the better. Thus in most Pareto
optimization diagrams you come across, the front will be those individuals closer to the origin.
I’ve strived to be consistent throughout this text, and so in this section we’ll continue to assume
that larger objective values are superior. Hence the organization of figures and algorithms in this
chapter.
7.1 Naive Methods

Before we get to the Pareto methods, let’s start with the more naive (but sometimes pretty good)
methods used to shoehorn multiobjective problems into a style usable by most “traditional” meta-
heuristic algorithms.
The simplest way to do this is to bundle all the objectives into a single fitness using some kind
of linear function. For example, maybe you feel that one unit of Cheap is worth ten units of Tall,
five units of Earthquake Resistant, and four units of Energy Efficient. Thus we might define the
quality of a solution as a weighted sum of how well it met various objectives:
1 1 1
Fitness(i ) = Cheapness(i ) + Height(i ) + EarthquakeResistance(i ) + EnergyEfficiency(i )
10 5 4
128
We’ve seen this theme a number of times in the past so far. For True Pareto Front
example: linear parsimony pressure; and the average of various test (Theoretical Optimum)
cases. There are three problems with this. First, you’re required to
come up with the degree to which one objective is worth another 6
objective. This is likely hard to do, and may be close to impossible B
if your objectives are nonlinear (that is, the difference between 9

and 10 height is much greater than the difference between 2 and 3 24
height, say). It’s the same basic problem discussed regarding linear
parsimony pressure in the Section 4.6 (Bloat). Second, realize that if M
20 A
Pareto dominates N, it’s already the case that Fitness(M) ≥ Fitness(N),
assuming your weights are positive. So a Pareto method in some 5
sense gives you some of this stuff for free already. Third, a weighted
sum may not match the goal of moving towards the Pareto Front. Figure 49 A may be considered
Consider the simplest scenario, where we’re adding up the objectives superior, but B has a higher total.
(that is, all weights are 1). We have two objectives, and Figure 49
shows the true Pareto front. Individual A is very close to the front, and so is the more desirable
individual. But Individual B sums to a higher value, and so would be selected over A using this
fitness strategy.
To solve the first problem (having to come up with weights), we could instead abandon linear
functions and simply treat the objectives as uncomparable functions. For example, perhaps we
simply invent preferences among the objectives in order to perform a lexicographic ordering: M
is better than N if it is superior in Height. If they’re the same Height, it’s better if it’s superior
in Cheapness. Then Earthquake Resistance. Then Energy Efficiency. We can provide a selection
procedure by extending Algorithm 63 (Lexicographic Tournament Selection) to the case of more
than two objectives. Basically when comparing two individuals, we run through the objectives
(most important to least important) until we find one clearly superior to the other in that objective.
Assuming we have an ObjectiveValue(objective, individual) function which tells us the quality of
individual with regard to the given objective, we might perform a tournament selection like this:
Algorithm 94 Multiobjective Lexicographic Tournament Selection

1: Best ← individual picked at random from population with replacement
2: O ← {O1 , ..., On } objectives to assess with " In lexicographic order, most to least preferred.
5: Next ← individual picked at random from population with replacement
6: for j from 1 to n do
7: if ObjectiveValue(O j , Next) > ObjectiveValue(O j , Best) then " Clearly superior
8: Best ← Next
9: break from inner for
10: else if ObjectiveValue(O j , Next) < ObjectiveValue(O j , Best) then " Clearly inferior
12: return Best
129
We could also pick an objective at random each time to use for fitness for this selection only:
Algorithm 95 Multiobjective Ratio Tournament Selection

2: O ← {O1 , ..., On } objectives to assess with
4: j ← random number picked uniformly from 1 to n

7: if ObjectiveValue(O j , Next) > ObjectiveValue(O j , Best) then
8: Best ← Next
9: return Best
Or we could use voting: an individual is preferred if it is ahead in more objectives:
Algorithm 96 Multiobjective Majority Tournament Selection

2: O ← {O1 , ..., On } objectives to assess with, more important objectives first
6: c←0
7: for each objective O j ∈ O do
8: if ObjectiveValue(O j , Next) > ObjectiveValue(O j , Best) then
9: c ← c+1
10: else if ObjectiveValue(O j , Next) < ObjectiveValue(O j , Best) then
11: c ← c−1
12: if c > 0 then
13: Best ← Next
14: return Best
Finally, we could extend Algorithm 64 (Double Tournament Selection) to the case of more
than two objectives. Here we perform a tournament based on one objective. The entrants to that
tournament are selected using tournament selections on a second objective. The entrants to that
tournament are selected using tournament selections on a third objective, and so on. Thus the
winner is more often that not a jack-of-all-trades which is pretty good in all objectives.
130
Algorithm 97 Multiple Tournament Selection
2: T ← { T1 , ..., Tn } tournament sizes for the objectives in O, all ≥ 1 " Allows different weights
3: return ObjectiveTournament(O, T)
4: procedure ObjectiveTournament(O, T)
6: n ← ||O|| " O and T change in size. The current last elements are On and Tn
7: if O − {On } is empty then " On is the last remaining objective!
9: else
10: Best ← ObjectiveTournament(O − {On }, T − { Tn }) " Delete the current objective
11: for i from 2 to Tn do
12: if O − {On } is empty then " This is the remaining objective!
14: else
15: Next ← ObjectiveTournament(O − {On }, T − { Tn }) " Delete the current objective
16: if ObjectiveValue(On , Next) > ObjectiveValue(On , Best) then
17: Best ← Next
18: return Best
7.2 Non-Dominated Sorting

The previous algorithms attempt to merge objectives into one single fitness value by trading off
one objective for another in some way. But a lot of current algorithms instead use notions of Pareto
domination to get a little more closely at what “better” means in a multiobjective sense.
One simple way to do this is to construct a tournament selection operator based on Pareto
domination. But first, let’s review the definition. Individual A Pareto dominates individual B if
A is at least as good as B in every objective and better than B in at least one objective. Here’s an
algorithm which computes that:
Algorithm 98 Pareto Domination
1: A ← individual A " We’ll determine: does A dominate B?
2: B ← individual B
4: a ← false " A is superior to B somewhere

5: for each objective Oi ∈ O do
6: if ObjectiveValue(Oi , A) > ObjectiveValue(Oi , B) then
7: a ← true
8: else if ObjectiveValue(Oi , B) > ObjectiveValue(Oi , A) then
9: return false
10: return a
131
Now we can build a binary tournament selection procedure based on Pareto domination:
Algorithm 99 Pareto Domination Binary Tournament Selection

1: P ← population
2: Pa ← individual picked at random from P with replacement

3: Pb ← individual picked at random from P with replacement
4: if Pa Pareto Dominates Pb then
Cheaper
5: return Pa
Rank 1
6: else if Pb Pareto Dominates Pa then Rank 2
7: return Pb
8: else
9: return either Pa or Pb , chosen at random Rank 4
Unfortunately, even if two individuals don’t Pareto- Rank 3

dominate one another, and thus are equally attractive to Rank 5
the experimenter, one individual might still be preferred More Energy Efficient
for optimization purposes. Specifically, if A has many in- Figure 50 Pareto ranks.
dividuals in the population who Pareto-dominate it, and B
has none, then we’re interested in selecting B because we’ll probably select individuals better than
A in the next generation anyway. Sure, B doesn’t Pareto dominate A. But A is part of the rabble.
To get at this notion, we need a notion of how close an individual is to the Pareto front. There
are various ways to do this, and we’ll discuss additional one (strength) in the next section. But we
start here with a new concept called a Pareto Front Rank. Individuals in the Pareto front are in
Rank 1. If we removed these individuals from the population, then computed a new front, individuals in
that front would be in Rank 2. If we removed those individuals, then computed a new front, we’d
get Rank 3, and so on. It’s like peeling an onion. Figure 50 shows the notion of ranks.
Let’s start by defining how to compute a Pareto front. The trick is to go through the population
and add an individual to the front if it isn’t dominated by anyone presently in the front, and remove
individuals from the front if they got dominated by this new individual. It’s fairly straightforward:
Algorithm 100 Computing a Pareto Non-Dominated Front

1: G ← { G1 , ..., Gm } Group of individuals to compute the front among " Often the population
3: F ← {} " The front

4: for each individual Gi ∈ G do
5: F ← F ∪ { Gi } " Assume Gi ’s gonna be in the front
6: for each individual Fj ∈ F other than Gi do
7: if Fj Pareto Dominates Gi given O then " Oh well, guess it’s not gonna stay in the front
8: F ← F − { Gi }
9: break out of inner for-loop
10: else if Gi Pareto Dominates Fj given O then " An existing front member knocked out!
11: F ← F − { Fj }
12: return F
132
Computing the ranks is easy: figure out the first front, then remove the individuals, then figure
out the front again, and so on. If we pre-process all the individuals with this procedure, we could
then simply use the Pareto Front Rank of an individual as its fitness. Since lower Ranks are better,
we could convert it into a fitness like this:
1
Fitness(i ) =
1 + ParetoFrontRank(i )
The algorithm to compute the ranks builds two results at once: first it partitions the population
P into ranks, with each rank (a group of individuals) stored in the vector F. Second, it assigns a
rank number to an individual (perhaps the individual gets it written internally somewhere). That
way later on we can ask both: (1) which individuals are in rank i, and (2) what rank is individual j
in? This procedure is called Non-Dominated Sorting, by N. Srinvas and Kalyanmoy Deb.106
Algorithm 101 Front Rank Assignment by Non-Dominated Sorting

1: P ← Population
3: P) ← P " We’ll gradually remove individuals from P)

4: R←$ % " Initially empty ordered vector of Pareto Front Ranks
5: i←1
6: repeat
7: Ri ← Pareto Non-Dominated Front of P) using O
8: for each individual A ∈ Ri do
9: ParetoFrontRank(A) ← i
10: P) ← P) − { A} " Remove the current front from P)
11: i ← i+1
12: until P) is empty
13: return R B1
Cheaper
B2
B A2
Sparsity We’d also like to push the individuals in the pop-

ulation towards being spread more evenly across the front.
A1
To do this we could assign a distance measure of some sort A
among individuals in the same Pareto Front Rank. Let’s

define the sparsity of an individual: an individual is in a
more sparse region if the closest individuals on either side More Energy Efficient
of it in its Pareto Front Rank aren’t too close to it.
Figure 51 illustrates the notion we’re more or less after. Figure 51 The sparsity of individual B is
higher than individual A because A1 + A2 <
We’ll define sparsity as Manhattan distance,107 over every B + B .
1 2
objective, between an individual’s left and right neighbors
106 First
published in N. Srinivas and Kalyanmoy Deb, 1994, Multiobjective optimization using nondominated sorting
in genetic algorithms, Evolutionary Computation, 2, 221–248. This paper also introduced Algorithm 100.
107 Manhattan lies on a grid, so you can’t go directly from point A to point B unless you’re capable of leaping tall
buildings in a single bound. Instead you must walk horizontally so many blocks, then vertically so many blocks. That’s
the Manhattan distance from A to B.
133
along its Pareto Front Rank. Individuals at the far ends of the Pareto Front Rank will be assigned
an infinite sparsity. To compute sparsity, you’ll likely need to know the range of possible values
that any given objective can take on (from min to max). If you don’t know this, you may be forced
to assume that the range equals 1 for all objectives.
Algorithm 102 Multiobjective Sparsity Assignment

1: R ← $...% provided Pareto Front Ranks of Individuals " To compute a single rank G, pass in $ G %.
3: Range(Oi ) function providing the range (max − min) of possible values for a given objective Oi
4: for each Pareto Front Rank F ∈ R do

5: for each individual Fj ∈ F do
6: Sparsity(Fj ) ← 0
7: for each objective Oi ∈ O do
8: F ) ← F sorted by ObjectiveValue given objective Oi
9: Sparsity(F1) ) ← ∞
10: Sparsity(F||) F|| ) ← ∞ " Each end is really really sparse!
11: )
for j from 2 to || F || − 1 do
) )
ObjectiveValue(Oi , Fj)+1 ) − ObjectiveValue(Oi , Fj)−1 )
12: Sparsity(Fj ) ← Sparsity(Fj ) +
Range(Oi )
13: return R with Sparsities assigned
We can now use sparsity to do a kind of crowding, but one which is in the multiobjective space
rather than in a genotype or phenotype space. We define a tournament selection to select first based
on Pareto Front Rank, but to break ties by using sparsity. The idea is to get individuals which are
not only close to the true Pareto front, but also nicely spread out along it.
Algorithm 103 Non-Dominated Sorting Lexicographic Tournament Selection With Sparsity

1: P ← Population with Pareto Front Ranks assigned
6: if ParetoFrontRank(Next) < ParetoFrontRank(Best) then " Lower ranks are better
7: Best ← Next
8: else if ParetoFrontRank(Next) = ParetoFrontRank(Best) then
9: if Sparsity(Next) > Sparsity(Best) then
10: Best ← Next " Higher sparsities are better
11: return Best
134
This alone does a pretty decent job. But the Non-Dominated Sorting Genetic Algorithm II (or
NSGA-II, by Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and T. Meyarivan)108 goes a little
bit further: it also keeps around all the best known individuals so far, in a sort of (µ + λ) or elitist
notion.
Algorithm 104 An Abstract Version of the Non-Dominated Sorting Genetic Algorithm II (NSGA-II)
1: P ← { P1 , ..., Pm } Build Initial Population
2: AssessFitness(P) " Compute the objective values for the Pareto front ranks
3: R ← $...% Pareto Front Ranks of P
4: for each Pareto Front Rank in Ri ∈ R do
5: Compute Sparsities of Individuals in Ri
6: BestFront ← Pareto Front of P
7: repeat
8: Q ← Breed(P), using Algorithm 103 for selection (typically with tournament size of 2)
9: AssessFitness(Q) " Compute the objective values for the Pareto front ranks
10: Q ← Q∪P
11: P ← {}
12: R ← Compute Front Ranks of Q
13: BestFront ← Pareto Front of Q
14: for each Front Rank Ri ∈ R do
15: Compute Sparsities of Individuals in $ Ri % " Just for Ri , no need for others
16: if || P|| + || Ri || ≥ m then " This will be our last front to load into P
17: P ← P ∪ the Sparsest m − || P|| individuals in Ri , breaking ties arbitrarily
18: break from the for loop
19: else
20: P ← P ∪ Ri " Just dump it in
21: until BestFront is the ideal Pareto front or we have run out of time
22: return BestFront
The general idea is to hold in P an archive of the best n individuals discovered so far. We then
breed a new population Q from P, and everybody in P and Q gets to compete for who gets to
stay in the archive. Such algorithms are sometimes known as archive algorithms. Ordinarily an
approach like this would be considered highly exploitative. But in multiobjective optimization
things are a little different because we’re not looking for just a single point in space. Instead we’re
looking for an entire Pareto front which is spread throughout the space, and that front alone imposes
a bit of exploration on the problem.
Note that we only compute Sparsities for a select collection of Pareto Front Ranks. This is
because they’re the only ones that ever use them: the other ranks get thrown away. You can just
compute Sparsities for all of Q if you want to, it’s no big deal.
108 Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and T. Meyarivan, 2000, A fast elitist non-dominated sorting
genetic algorithm for multi-objective optimization: NSGA-II, in Marc Schoenauer, et al., editors, Parallel Problem Solving
from Nature (PPSN VI), pages 849–858, Springer. This paper also introduced Algorithm 102.
135
7.3 Pareto Strength
Pareto Front Ranks are not the only way we can use Pareto values to compute fitness. We could
also identify the strength of an individual, defined as the number of individuals in the population
that the individual Pareto dominates.
We could use an individual’s strength as his fitness. There’s a problem with this, however.
Strength doesn’t necessarily correspond with how close an individual is to the Pareto front. Indeed,
individuals near the corners of the front are likely to not be very strong compared to individuals
fairly distant from the front, as shown in Figure 52. Alternatively, we may define the weakness of
an individual to be the number of individuals which dominate the individual. Obviously individuals
on the Pareto front have a 0 weakness, and individuals far from the front are likely to have a high
weakness. A slightly more refined version of weakness is the wimpiness109 of an individual: the
sum total strength of everyone who dominates the individual, for an individual i and a group G, that is,
Wimpiness(i ) = ∑ Strength( g)
g∈ G that Pareto Dominate i
Ideally we’d like the wimpiness of an individual to be as low as possible. A non-dominated

individual has a 0 wimpiness. We could use some kind of non-wimpiness as a fitness too. To do
this, we could convert wimpiness such that wimpier individuals have lower values. Perhaps:
1
Fitness(i ) =
1 + Wimpiness(i )
Eckart Zitzler, Marco Laumanns, and Lothar Thiele

built an archive-based algorithm around the notion
Cheaper
of strength (or more correctly, wimpiness), called the

Strength Pareto Evolutionary Algorithm (or SPEA). The B
current version, SPEA2, competes directly with NSGA-II
and various other multiobjective stochastic optimization
algorithms.110 Like NSGA-II, SPEA2 maintains an archive
of the best known Pareto front members plus some oth- A
ers. SPEA2 also similarly employs both a Pareto measure
and a crowding measure in its fitness procedure. However, More Energy Efficient
SPEA2’s Pareto measure is Wimpiness, and its crowding
measure is based on distance to other individuals in the Figure 52 Individual A is closer to the Pareto
front, but individual B is Stronger.
multiobjective space, rather than distance along ranks.
SPEA2’s similarity measure computes a distance to
other individuals in the population, and specifically, to the kth closest individual in the popu-
lation. There are many fancy ways of computing this in a reasonably efficient manner. Here I’m
just going to suggest a grotesquely inefficient, but simple, approach.111 Basically we compute
109 Ofcourse I made up these names (except for strength).
110 They’re sort of intertwined. SPEA was introduced in Eckart Zitzler and Lothar Thiele, 1999, Multiobjective evo-
lutionary algorithms: A comparitive case study and the strength pareto approach, IEEE Transactions on Evolutionary
Computation, 3(4), 257–271. NSGA-II then came out in 2000, and SPEA2 then came out as Eckart Zitzler, Marco Laumanns,
and Lothar Thiele, 2002, SPEA2: Improving the strength pareto evolutionary algorithm for multiobjective optimization,
in K. Giannakoglou, et al., editors, Evolutionary Methods for Design, Optimization, and Control, pages 19–26.
111 Hey, fitness assessment time is the dominant factor timewise nowadays anyway!
136
the distance from everyone to everyone. Then, for each individual in the population, we sort the
population by distance to that individual, and take the kth closest individual. This is O(n2 lg n),
where n is the population size. That’s not great.
Algorithm 105 Compute the Distance of the Kth Closest Individual

1: P ← { P1 , ..., Pm } Population
3: Pl ← individual to compute kth closest individual to
4: k ← desired individual index (the kth individual from l)
5: global D ← m vectors, each of size m " Di holds a vector of distances of various individuals i
6: global S ← {S1 , ..., Sm } " Si will be true if Di has already been sorted
7: perform once only
9: V ← {} " Our distances
10: for each individual Pj ∈ P do
.' /
n 2
11: V ←V∪ ∑ m =1 ( ObjectiveValue (O ,
m iP ) − ObjectiveValue (O ,
m jP ))
12: Di ← V
13: Si ← false
15: if Sl is false then " Need to sort
16: Sort Dl , smallest first
17: Sl ← true
18: W ← Dl
19: return Wk+1 " It’s Wk+1 because W1 is always 0: the distance to the same individual
Given the Wimpiness of an individual and the kth closest individual to it, we can finally define
a fitness. Define a pre-fitness value Gi as follows:
1
Gi ← Wimpiness(i ) +
2 + di
0) 1
... where di is the distance to the kth closest individual to i, where k = || P|| typically.112 The
smaller the value of Gi the better. The idea is that a big distance di makes Gi smaller (because it’s
far away from other individuals — we want diversity!) and likewise a small Wimpiness makes Gi
smaller.
SPEA2 in reality uses Gi as the fitness of individual i: but in keeping with our tradition (higher
fitness is better), let’s convert it into a final fitness like we’ve done before:
1
Fitness(i ) =
1 + Gi
2) 3
112 Actually, Zitzler and Thiele don’t say how you should round it: you could just as well do k = || P|| I suppose.
137
Each iteration, SPEA2 will build an archive consisting of the current Pareto front of the popula-
tion. The archive is supposed to be of size n. If there aren’t enough individuals in the front to fill all
those n, SPEA2 will fill the rest with other fit individuals selected from the population. If there are
instead too many individuals in the Pareto front to fit into n, SPEA2 needs to trim some individuals.
It does this by iteratively deleting individuals who have the smallest kth closest distance (starting
with k = 1, breaking ties with k = 2, and so on). The goal is to get in the archive those individuals in
the Pareto front which are furthest away from one another and other individuals in the population.
The algorithm for constructing the archive looks like this:
Algorithm 106 SPEA2 Archive Construction

1: P ← { P1 , ..., Pm } Population
3: a ← desired archive size
4: A ← Pareto non-dominated front of P " The archive

5: Q ← P−A " All individuals not in the front
6: if || A|| < a then " Too small! Pack with some more individuals
7: Sort Q by fitness
8: A ← A ∪ the a − || A|| fittest individuals in Q, breaking ties arbitrarily
9: while || A|| > a do " Too big! Remove some “k-closest” individuals
10: Closest ← A1
11: c ← index of A1 in P
12: for each individual Ai ∈ A except A1 do
13: l ← index of Ai in P
14: for k from 1 to m − 1 do " Start with k = 1, break ties with larger values of k
15: if DistanceOfKthNearest(k, Pl ) < DistanceOfKthNearest(k, Pc ) then
16: Closest ← Ai
17: c←l
19: else if DistanceOfKthNearest(k, Pl ) > DistanceOfKthNearest(k, Pc ) then
21: A ← A − {Closest}
22: return A
Now we’re ready to describe the SPEA2 top-level algorithm. It’s very simple actually. Given a
population P and an (initially empty) archive A, we build a new archive of the Pareto Front from
P ∪ A, trimmed if necessary of “close” individuals, plus some other fit individuals from P to fill in
any gaps. Then we create a new population P by breeding from A (which eventually comes close to
random selection as the Pareto front improves). Does this sound like a bit like (µ + λ)? It should!
138
Algorithm 107 An Abstract Version of the Strength Pareto Evolutionary Algorithm 2 (SPEA2)
1: m ← desired population size
2: a ← desired archive size " Typically a = m
3: P ← { P1 , ..., Pm } Build Initial Population

4: A ← {}
5: repeat
6: AssessFitness(P)
7: BestFront ← Pareto Front of P ∪ A
8: A ← Construct SPEA2 Archive of size a from P ∪ A
9: P ← Breed(A), using tournament selection of size 2 " Fill up to the old size of P
10: until BestFront is the ideal Pareto front or we have run out of time
11: return BestFront
SPEA2 and NSGA-II both are basically versions of (µ + λ) in multiobjective space, coupled with
a diversity mechanism and a procedure for selecting individuals that are closer to the Pareto front.
Both SPEA2 and NSGA-II are fairly impressive algorithms,113 though NSGA-II is a bit simpler and
has lower computational complexity in unsophisticated versions.
113 Believe me, I know. Zbigniew Skolicki and I once constructed a massively parallel island model for doing multiob-
jective optimization. If there were n objectives, the islands were organized in a grid with n corners, one per objective.
For example with 2 objectives, the grid was a line. If there were 3 objectives, the grid was a triangle mesh. If there were 4
objectives, the grid was a mesh filling the volume of a tetrahedron (three-sided pyramid). Each island assessed fitness as
a weighted sum of the objectives. The closer an island was to a corner, the more it weighted that corner’s objective. Thus
islands in the corners or ends were 100% a certain objective, while (for example) islands near the center weighted each
objective evenly. Basically each island was searching for its own part of the Pareto front, resulting in (hopefully) a nicely
distributed set of points along the front. We got okay results. But SPEA2, on a single machine, beat our pants off.
139
140
8 Combinatorial Optimization
So far the kinds of problems we’ve tack-
led are very general: any arbitrary search
1", $3
space. We’ve seen spaces in the forms of per- 5/4", $4
mutations of variables (fixed-length vectors); 2", $11

spaces that have reasonable distance metrics
defined for them; and even spaces of trees or 1/5", $5 2/3", $1
sets of rules.
One particular kind of space deserves 4/5", $9
1/2", $13 157/60"
special consideration. A combinatorial op- 1/3", $7
timization problem114 is one in which the

solution consists of a combination of unique Available Blocks Knapsack
(Shown with Heights and Values) (Shown with Height)
components selected from a typically finite,
and often small, set. The objective is to find Figure 53 A knapsack problem. Fill the knapsack with as
the optimal combination of components. much value ($$$) without exceeding the knapsack’s height.
A classic combinatorial optimization
problem is a simple form of the knapsack problem: we’re given n blocks of different heights
and worth different amounts of money (unrelated to the heights) and a knapsack115 of a certain
larger height, as shown in Figure 53. The objective is to fill the knapsack with blocks worth the most
$$$ (or eee or ¥¥¥) without overfilling the knapsack.116 Blocks are the components. Figure 54
shows various combinations of blocks in the knapsack. As you can see, just because the knapsack
is maximally filled doesn’t mean it’s optimal: what counts is how much value can be packed into
the knapsack without going over. Overfull solutions are infeasible (or illegal or invalid).
This isn’t a trivial or obscure problem. It’s
got a lot of literature behind it. And lots
4/5", $9
of real-world problems can be cast into this 1/2", $13 1/3", $7
framework: knapsack problems show up in 1/3", $7
the processor queues of operating systems; 2/3", $1 2/3", $1
in allocations of delivery trucks along routes; 5/4", $4
1/5", $5
1/5", $5
and in determining how to get exactly $15.05
worth of appetizers in a restaurant.117 2", $11 4/5", $9
Another example is the classic travel- 1", $3

5/4", $4
ing salesman problem (or TSP), which has 1/2", $13
a set of cities with some number of routes

(plane flights, say) between various pairs Underfull Overfull 100% Filled Not 100% Filled
cities. Each route has a cost. The salesman (Not Legal) But only $23 But Optimal ($35)
must construct a tour starting at city A, vis- Figure 54 Filling the knapsack.
iting all the cities at least once, and finally
114 Not to be confused with combinatorics, an overall field of problems which could reasonably include, as a small subset,
practically everything discussed so far.
115 Related are various bin packing problems, where the objective is to figure out how to arrange blocks so that they
will fit correctly in a multi-dimensional bin.

116 There are various knapsack problems. For example, another version allows you to have as many copies of a given
block size as you need.

117 http://xkcd.com/287/
141
returning to A. Crucially, this tour must have the lowest cost possible. Put another way, the cities
are nodes and the routes are edges in a graph, labelled by cost, and the object is to find a minimum-
cost cycle which visits every node at least once. Here the components aren’t blocks but are rather
the edges in the graph. And the arrangement of these edges matters: there are lots of sets of edges
which are nonsense because they don’t form a cycle.
Costs and Values While the TSP has cost (the edge weights) which must be minimized, Knapsack
instead has value ($$$) which must be maximized. These are really just the same thing: simply
negate or invert the costs to create values. Most combinatorial optimization algorithms traditionally
assume costs, but we’ll include both cases. At any rate, one of many ways you might convert the
cost of a component Ci into a value (or vice versa) would be something along the lines of:
1
Value(Ci ) =
Cost(Ci )
That’s the relationship we’ll assume in this Section. This of course assumes that your costs (and
values) are > 0, which is the usual case. If your costs or values are both positive and negative,
some of the upcoming methods do a kind value-proportional selection, so you’ll need to add some
amount to make them all positive. Finally, there exist problems in which components all have
exactly the same value or cost. Or perhaps you might be able to provide your algorithm with a
heuristic118 that you as a user have designed to favor certain components over others. In this case
you could use Value(Ci ) = Heuristic(Ci ).
Knapsack does have one thing the TSP doesn’t have: it has additional weights119 (the block
heights) and a maximum “weight” which must not be exceeded. The TSP has a different notion of
infeasible solutions than simply ones which exceed a certain bound.
8.1 General-Purpose Optimization and Hard Constraints

Combinatorial optimization problems can be solved by most general-purpose metaheuristics such
as those we’ve seen so far, and in fact certain techniques (Iterated Local Search, Tabu Search, etc.)
are commonly promoted as combinatorics problem methods. But some care must be taken because
most metaheuristics are really designed to search much more general, wide-open spaces than the
constrained ones found in most combinatorial optimization problems. We can adapt them but need
to take into consideration these restrictions special to these kinds of problems.120
As an example, consider the use of a boolean vector in combination with a metaheuristic such
as simulated annealing or the genetic algorithm. Each slot in the vector represents a component,
and if the slot is true, then the component is used in the candidate solution. For example, in
Figure 53 we have blocks of height 2, 13 , 54 , 15 , 45 , 1, 23 , and 12 . A candidate solution to the problem
in this Figure would be a vector of eight slots. The optimal answer shown in Figure 54 would be
$false, true, false, true, true, false, true, true%, representing the blocks 13 , 15 , 45 , 23 , and 12 .
The problem with this approach is that it’s easy to create solutions which are infeasible. In
the knapsack problem we have declared that solutions which are larger than the knapsack are
118 A heuristic is a rule of thumb provided by you to the algorithm. It can often be wrong, but is right often enough that
it’s useful as a guide.
119 Yeah, confusing. TSP edge weights vs. combinatorial component weights. That’s just the terminology, sorry.
120 A good overview article on the topic, by two greats in the field, is Zbigniew Michalewicz and Marc Schoenauer,
1996, Evolutionary algorithms for constrained parameter optimization problems, Evolutionary Computation, 4(1), 1–32.
142
simply illegal. In Knapsack, it’s not a disaster to have candidate solutions like that, as long as
the final solution is feasible — we could just declare the quality of such infeasible solutions to be
their distance from the optimum (in this case perhaps how overfull the knapsack is). We might
punish them further for being infeasible. But in a problem like the Traveling Salesman Problem,
our boolean vector might consist of one slot per edge in the TSP graph. It’s easy to create infeasible
solutions for the TSP which are simply nonsense: how do we assess the “quality” of a candidate
solution whose TSP solution isn’t even a tour?
The issue here is that these kind of problems, as configured, have hard constraints: there are
large regions in the search space which are simply invalid. Ultimately we want a solution which is
feasible; and during the search process it’d be nice to have feasible candidate solutions so we can
actually think of a way to assign them quality assessments! There are two parts to this: initialization
(construction) of a candidate solution from scratch, and Tweaking a candidate solution into a new
one.
Construction Iterative construction of components within hard constraints is sometimes straight-

forward and sometimes not. Often it’s done like this:
1. Choose a component. For example, in the TSP, pick an edge between two cities A and B. In
Knapsack, it’s an initial block. Let our current (partial) solution start with just that component.
2. Identify the subset of components that can be concatenated to components in our partial
solution. In the TSP, this might be the set of all edges going out of A or B. In Knapsack, this is
all blocks that can still be added into the knapsack without going over.
3. Tend to discard the less desirable components. In the TSP, we might emphasize edges that
are going to cities we’ve not visited yet if possible.
4. Add to the partial solution a component chosen from among those components not yet
discarded.
5. Quit when there are no components left to add. Else go to step 2.
This is an intentionally vague description because iterative construction is almost always highly
problem-specific and often requires a lot of thought.
Tweaking The Tweak operator can be even harder to do right, because in the solution space
feasible solutions may be surrounded on all sides by infeasible ones. Four common approaches:
• Invent a closed Tweak operator which automatically creates feasible children. This can be a
challenge to do, particularly if you’re including crossover. And if you create a closed operator,
can it generate all possible feasible children? Is there a bias? Do you know what it is?
• Repeatedly try various Tweaks until you create a child which is feasible. This is relatively
easy to do, but it may be computationally expensive.
• Allow infeasible solutions but construct a quality assessment function for them based on
their distance to the nearest feasible solution or to the optimum. This is easier to do for some
problems than others. For example, in the Knapsack problem it’s easy: the quality of an
overfull solution could be simply based on how overfull it is (just like underfull solutions).
143
• Assign infeasible solutions a poor quality. This essentially eliminates them from the popula-
tion; but of course it makes your effective population size that much smaller. It has another
problem too: moving just over the edge between the feasible and infeasible regions in the
space results in a huge decrease in quality: it’s a Hamming Cliff (see Representation, Section
4). In Knapsack, for example, the best solutions are very close to infeasible ones because
they’re close to filled. So one little mutation near the best solutions and whammo, you’re
infeasible and have big quality punishment. This makes optimizing near the best solutions a
bit like walking on a tightrope.
None of these is particularly inviting. While it’s often easy to create a valid construction
operator, making a good Tweak operator that’s closed can be pretty hard. And the other methods
are expensive or allow infeasible solutions in your population.
Component-Oriented Methods The rest of this Section concerns itself with methods specially
designed for certain kinds of spaces often found in combinatorial optimization, by taking advantage
of the fact that the that solutions in these spaces consist of combinations of components drawn from a
typically fixed set. It’s the presence of this fixed set that we can take advantage of in a greedy, local
fashion by maintaining historical “quality” values, so to speak, of individual components rather
than (or in addition to) complete solutions. There are two reasons you might want to do this:
• While constructing, to tend to select from components which have proven to be better choices.
• While Tweaking, to modify those components which appear to be getting us in a local

optimum.
We’ll begin with a straightforward metaheuristic called Greedy Randomized Adaptive Search
Procedures (or GRASP) which embodies the basic notion of constructing combinatorial solutions
out of components, then Tweaking them. From there we will move to a related technique, Ant
Colony Optimization, which assigns “historical quality” values to these components to more
aggressively construct solutions from the historically “better” components. Finally, we’ll examine a
variation of Tabu Search called Guided Local Search which focuses instead on the Tweak side of
things: it’s designed to temporarily “punish” those components which have gotten the algorithm
into a rut.
Some of these methods take advantage of the “historical quality” values of individual com-
ponents, but use them in quite different ways. Ant Colony Optimization tries to favor the best-
performing components; but Guided Local Search gathers this information to determine which
low-performing components appear to show up often in local optima.
The meaning of Quality or Fitness Because combinatorial problems can be cast as either cost
or as value, the meaning of quality or fitness of a candidate solution is shaky. If your problem is
in terms of value (such as Knapsack), it’s easy to define quality or fitness simply as the sum total
value, that is, ∑i Value(Ci ), of all the components Ci which appear in the candidate solution. If
your problem is in terms of cost (such as the TSP), it’s not so easy: you want the presence of many
low-cost components to collectively result in a high-quality solution. A common approach is to
define quality or fitness as 1/(∑i Cost(Ci )), for each component Ci that appears in the solution.
144
8.2 Greedy Randomized Adaptive Search Procedures
At any rate, let’s start easy with a single-state metaheuristic which is built on the notions of
constructing and Tweaking feasible solutions, but which doesn’t use any notion of component-level
“historical quality”: Greedy Randomized Adaptive Search Procedures or GRASP, by Thomas Feo
and Mauricio Resende.121 The overall algorithm is really simple: we create a feasible solution by
constructing from among highest value (lowest cost) components (basically using the approach
outlined earlier) and then do some hill-climbing on the solution.
Algorithm 108 Greedy Randomized Adaptive Search Procedures (GRASP)

1: C ← {C1 , ..., Cn } components
2: p ← percentage of components to include each iteration
3: m ← length of time to do hill-climbing
4: Best ← !
5: repeat
6: S ← {} " Our candidate solution
7: repeat
8: C ) ← components in C − S which could be added to S without being infeasible
9: if C ) is empty then
10: S ← {} " Try again
11: else
12: C )) ← the p% highest value (or lowest cost) components in C )
13: S ← S ∪ {component chosen uniformly at random from C )) }
14: until S is a complete solution
15: for m times do
16: R ← Tweak(Copy(S)) " Tweak must be closed, that is, it must create feasible solutions
18: S←R
19: if Best = ! or Quality(S) > Quality(Best) then
20: Best ← S
22: return Best
Instead of picking the p% best available components, some versions of GRASP pick components
from among the components whose value is no less than (or cost is no higher than) some amount.
GRASP is more or less using a truncation selection among components to do its initial construction
of candidate solutions. You could do something else like a tournament selection among the
components, or a fitness-proportionate selection procedure (see Section 3 for these methods).
GRASP illustrates one way how to construct candidate solutions by iteratively picking compo-
nents. But it’s still got the same conundrum that faces evolutionary computation when it comes to
the Tweak step: you have to come up with some way of guaranteeing closure.
121 The first GRASP paper was Thomas A. Feo and Mauricio G. C. Resende, 1989, A probabilistic heuristic for a compu-
tationally difficult set covering problem, Operations Research Letters, 8, 67–71. Many of Resende’s current publications on
GRASP may be found at http://www.research.att.com/∼mgcr/doc/
145
8.3 Ant Colony Optimization
Marco Dorigo’s Ant Colony Optimization (or ACO)122 is an approach to combinatorial optimization
which gets out of the issue of Tweaking by making it optional. Rather, it simply assembles candidate
solutions by selecting components which compete with one another for attention.
ACO is population-oriented. But there are two different kinds of “populations” in ACO. First,
there is the set of components that make up a candidate solutions to the problem. In the Knapsack
problem, this set would consist of all the blocks. In the TSP, it would consist of all the edges. The
set of components never changes: but we will adjust the “fitness” (called the pheromone) of the
various components in the population as time goes on.
Each generation we build one or more candidate solutions, called ant trails in ACO parlance, by
selecting components one by one based, in part, on their pheromones. This constitutes the second
“population” in ACO: the collection of trails. Then we assess the fitness of each trail. For each trail,
each of the components in that trail is then updated based on that fitness: a bit of the trail’s fitness
is rolled into each component’s pheromone. Does this sound like some kind of one-population
cooperative coevolution?
The basic abstract ACO algorithm:
Algorithm 109 An Abstract Ant Colony Optimization Algorithm (ACO)

2: popsize ← number of trails to build at once " “ant trails” is ACOspeak for “candidate solutions”
3: !p ← $ p1 , ..., pn % pheromones of the components, initially zero

4: Best ← !
5: repeat
6: P ← popsize trails built by iteratively selecting components based on pheromones and costs or
values
7: for Pi ∈ P do
8: Pi ← Optionally Hill-Climb Pi
10: Best ← Pi
11: Update !p for components based on the fitness results for each Pi ∈ P in which they participated
13: return Best
I set this up to highlight its similarities to GRASP: both algorithms iteratively build candidate
solutions, then hill-climb them. There are obvious differences though. First, ACO is builds some
popsize candidate solutions all at once. Second, ACO’s hill-climbing is optional, and indeed it’s often
not done at all. If you’re finding it difficult to construct a closed Tweak operator for your particular
representation, you can entirely skip the hill-climbing step if need be.
Third, and most importantly, components are selected not just based on component value or
cost, but also on pheromones. A pheromone is essentially the “historical quality” of a component:
122 ACO’s been around since around 1992, when it Dorigo proposed it in his dissertation: Marco Dorigo, 1992,
Optimization, Learning and Natural Algorithms, Ph.D. thesis, Politecnico di Milano, Milan, Italy. The algorithms here are
loosely adapted from Dorigo and Thomas Stützle’s excellent recent book: Marco Dorigo and Thomas Stützle, 2004, Ant
Colony Optimization, MIT Press.
146
often approximately the sum total (or mean, etc.) fitness of all the trails that the component has
been a part of. Pheromones tell us how good a component would be to select regardless of its
(possibly low) value or (high) cost. After assessing the fitness of trails, we update the pheromones
in some way to reflect new fitness values we’ve discovered so those components are more or less
likely to be selected in the future.
So where are the ants? Well, here’s the thing. ACO was inspired by earlier research work in
pheromone-based ant foraging and trail formation algorithms: but the relationship between ACO
and actual ants is... pretty thin. ACO practitioners like to weave the following tale: to solve the
Traveling Salesman Problem, we place an Ant in Seattle and tell it to go wander about the graph,
from city to city, eventually forming a cycle. The ant does so by picking edges (trips to other cities
from the ant’s current city) that presently have high pheromones and relatively good (low) edge
costs. After the ant has finished, it lays a fixed amount of pheromone on the trail. If the trail is
shorter (lower costs), then of course that pheromone will be distributed more densely among its
edges, making them more desirable for future ants.
That’s the story anyway. The truth is, there are no ants. There are just components with historical
qualities (“pheromones”), and candidate solutions formed from those components (the “trails”),
with fitness assessed to those candidate solutions and then divvied up among the components
forming them.
8.3.1 The Ant System

The first version of ACO was the Ant System or AS. It’s not used as often nowadays but is a good
starting point to illustrate these notions. In the Ant System, we select components based on a
fitness-proportionate selection procedure of sorts, employing both costs or values and pheromones
(we’ll get to that). We then always add fitnesses into the component pheromones. Since this could
cause the pheromones to go sky-high, we also always reduce (or evaporate) all pheromones a bit
each time.
The Ant System has five basic steps:
1. Construct some trails (candidate solutions) by selecting components.
2. (Optionally) Hill-Climb the trails to improve them.
3. Assess the fitness of the final trails.
4. “Evaporate” all the pheromones a bit.
5. Update the pheromones involved in trails based on the fitness of those solutions.
In the original AS algorithm, there’s no hill-climbing: I’ve added it here. Later versions of ACO
include it. Here’s a version of the algorithm (note certain similarities with GRASP):
147
Algorithm 110 The Ant System (AS)
2: e ← evaporation constant, 0 < e ≤ 1
3: popsize ← number of trails to construct at once
4: γ ← initial value for pheromones
5: t ← iterations to Hill-Climb
6: !p ← $ p1 , ..., pn % pheromones of the components, all set to γ

7: Best ← !
8: repeat
9: P ← {} " Our trails (candidate solutions)
10: for popsize times do " Build some trails
11: S ← {}
12: repeat
16: else
17: S ← S ∪ {component selected from C ) based on pheromones and values or costs}
18: until S is a complete trail
19: S ← Hill-Climb(S) for t iterations " Optional. By default, not done.
20: AssessFitness(S)
21: if Best = ! or Fitness(S) > Fitness(Best) then
22: Best ← S
23: P ← P ∪ {S}
24: for each pi ∈ !p do " Decrease all pheromones a bit (“evaporation”)
25: p i ← (1 − e ) p i
26: for each Pj ∈ P do " Update pheromones in components used in trails
27: for each component Ci do
28: if Ci was used in Pj then
29: pi ← pi + Fitness(Pj )
31: return Best
Component Values or Costs, and Selecting Components We construct trails by repeatedly se-
lecting from those components which, if added to the trail, wouldn’t make it infeasible. Knapcksack
is easy: keep on selecting blocks until it’s impossible to select one without going over. But the TSP
is more complicated. For example, in the TSP we could just keep selecting edges until we have
a complete tour. But we might wind up with edges we didn’t need, or a bafflingly complex tour.
Another approach might be to start with a city, then select from among those edges going out of
the city to some city we’ve not seen yet (unless we have no choice), then select from among edges
going out of that city, and so on. However it may be the case that the optimal tour requires that we
go through certain cities repeatedly. Or what if the only possible tours require that you go from
148
Salt Lake City to Denver, yet that’s got a high cost (low value) so we keep avoiding it and picking
other cities, only to be forced to backtrack? We could have some pretty ugly tours. Anyway: the
point is, trail construction can require some forethought.
AS selects components based on what I’ll call their desirability: a linear combination of value
and pheromones:
%
Desirability(Ci ) = piδ + (Value(Ci ))
...or if your problem is using costs...
4 5%
1
Desirability(Ci ) = piδ +
Cost(Ci )
δ and % are tuning parameters. Note that as the pheromone goes up the quality goes up.
Likewise, if a component has a higher value (or lower cost), then the quality goes up. Now AS
simply does a “desirability-proportionate” selection among the components we’re considering,
similar to Algorithm 30.
If you like you could perform some other selection procedure among your components, like
tournament selection or GRASP-style truncation to p% based on desirability.
Initializing the Pheromones You could set them all to γ = 1. For the TSP, the ACO folks often set
them to γ = popsize × (1/Cost( D )), where D is some costly, absurd tour like the Nearest Neighbor
Tour (construct a TSP tour greedily by always picking the lowest cost edge).
Evaporating Pheromones The Ant System evaporates pheromones because otherwise the
pheromones keep on piling up. But there’s perhaps a better way to do it: adjust the pheromones up
or down based on how well they’ve performed on average. Instead of evaporating and updating
as was shown in the Ant System, we could just take each pheromone pi and adjust it as follows:
Algorithm 111 Pheromone Updating with a Learning Rate

2: !p ← $ p1 , ..., pn % pheromones of the components
3: P ← { P1 , ..., Pm } population of trails
4: α ← learning rate
5: !r ← $r1 , ..., rn % total desirability of each component, initially 0

6: !c ← $c1 , ..., cn % component usage counts, initially 0
7: for each Pj ∈ P do " Compute the average fitness of trails which employed each component
8: for each component Ci do
9: if Ci was used in Pj then
10: ri ← ri + Desirability(Pj )
11: ci ← ci + 1
12: for each pi ∈ !p do
13: if ci > 0 then
14: pi ← (1 − α) pi + α crii " crii is the average fitness computed earlier
15: return !p
149
0 ≤ α ≤ 1 is the learning rate. For each component, we’re computing the average fitness of
every trail which used that component. Then we’re throwing out a small amount of what we know
so far (1 − α’s worth) and rolling in a little bit of what we’ve just learned this iteration about how
good a component is (α’s worth). If α is large, we quickly adopt new information at the expense of
our historical knowledge. It’s probably best if α is small.123
Optional Hill-Climbing: More Exploitation AS doesn’t have hill-climbing by default. But we

could hill-climb the ant trail S it right after the AssessFitness step, just like we do in GRASP. And
just like in GRASP we’re going to have the same issue: guaranteeing that each time we Tweak an
ant trail, the child is still a valid ant trail. For some problems this is easy, for others, not so easy.
Anyway, hill-climbing adds more exploitation to the problem, directly moving towards the locally
best solutions we can find. Often this is a good approach for problems like TSP, which tend to
benefit from a high dose of exploitation.
8.3.2 The Ant Colony System: A More Exploitative Algorithm

There have been a number of improvements on AS since it was first proposed (some of which were
mentioned earlier). Here I’ll mention one particularly well-known one: the Ant Colony System
(ACS).124 ACS works like the Ant System but with the following changes:
1. The use of an elitist approach to updating pheromones: only increase pheromones for
components used in the best trail discovered so far. In a sense this starts to approach (1 + λ).
2. The use of a learning rate in pheromone updates.
3. A slightly different approach for evaporating pheromones.
4. A strong tendency to select components that were used in the best trail discovered so far.
Elitism ACS only improves the pheromones of components that were used in the best-so-
far trail (the trail we store in Best), using the learning rate method stolen from Algorithm
111. That is, if a component is part of the best-so-far trail, we increase its pheromones as
pi ← (1 − α) pi + α Fitness(Best).
This is very strongly exploitative, so all pheromones are also decreased whenever they’re used
in a solution, notionally to make them less desirable for making future solutions in order to push
the system to explore a bit more in solution space. More specifically, whenever a component Ci is
used in a solution, we adjust its pheromone pi ← (1 − β) pi + βγ, where β is a sort of evaporation
or“unlearning rate”, and γ is the value we initialized the pheromones to originally. Left alone, this
would eventually reset the pheromones to all be γ.
Elitist Component Selection Component selection is also pretty exploitative. We flip a coin of
probability q. If it comes up heads, we select the component which has the highest Desirability.
Otherwise we select in the same way as AS selected: though ACS simplifies the selection mechanism
by getting rid of δ (setting it to 1).
123 We’ll see the 1 − α vs. α learning rate metaphor again in discussion of Learning Classifier Systems. It’s a common
notion in reinforcement learning too.

124 Again by Marco Dorigo and Luca Gambardella — no, there are plenty of people doing ACO besides Marco Dorigo!
150
Now we’re ready to do the Ant Colony System. It’s not all that different from AS in structure:
Algorithm 112 The Ant Colony System (ACS)

1: C ← {C1 , ..., Cn } Components
2: popsize ← number of trails to construct at once
3: α ← elitist learning rate
4: β ← evporation rate
5: γ ← initial value for pheromones
6: δ ← tuning parameter for heuristics in component selection " Usually δ = 1
7: % ← tuning parameter for pheromones in component selection
8: t ← iterations to Hill-Climb
9: q ← probability of selecting components in an elitist way
10: !p ← $ p1 , ..., pn % pheromones of the components, all set to γ

11: Best ← !
12: repeat
13: P ← {} " Our candidate solutions
14: for popsize times do " Build some trails
15: S ← {}
16: repeat
20: else
21: S ← S ∪ { component selected from C ) using Elitist Component Selection }
22: until S is a complete trail
23: S ← Hill-Climb(S) for t iterations " Optional. By default, not done.
26: Best ← S
27: for each pi ∈ !p do " Decrease all pheromones a bit (“evaporation”)
28: pi ← (1 − β) pi + βγ
29: for each component Si do " Update pheromones only of components in Best
30: if Si was used in Best then
31: pi ← (1 − α) pi + α Fitness(Best)
33: return Best
As before, we might be wise to do some hill-climbing right after the AssessFitness step.
At this point you may have picked up on an odd feature about ACO. The selection of com-
ponents in candidate solutions is greedily based on how well a component has appeared in
high-quality solutions (or perhaps even the best solution so far). It doesn’t consider the possibil-
ity that a component needs to always appear with some other component in order to be good,
and without the second component it’s terrible. That is, ACO completely disregards linkage among
components.
151
That’s a pretty bold assumption. This could, in theory, lead to the same problems that co-
operative coevolution has: jacks-of-all-trades. ACS tries to get around this by pushing hard for
the best-so-far result, just as cooperative coevolution’s best-of-n approaches and archive methods
try to view components in the light of their best situation. I think ACO has a lot in common
with coevolution, although it’s not been well studied. In some sense we may view ACO as a one
population pseudo-cooperative coevolution algorithm.
Perhaps it might be interesting to try ACO in a different way: a population not of components
but of (say) all possible pairs of components, for all possible pairs. We could select pairs that have
been performing well. This would move up the chain a little bit as far as linkage is concerned,
though it’d make a much bigger population.
ACO also has a lot in common with Univariate Estimation of Distribution Algorithms (dis-
cussed in Section 9.2)125 Here’s how to look at it: the components’ fitnesses may be viewed as
probabilities and the whole population is thus one probability distribution on a per-component
basis. Contrast this to the evolutionary model, where the population may also be viewed as a
sample distribution over the joint space of all possible candidate solutions, that is, all possible com-
binations of components. It should be obvious that ACO is searching a radically simpler (perhaps
simplistic) space compared to the evolutionary model. For general problems that may be an issue.
But for many combinatorial problems, it’s proven to be a good tradeoff.
8.4 Guided Local Search

There’s another way we can take advantage of the special component-based space found in
combinatorial optimization problems: by marking certain components which tend to cause local
optima and trying to avoid them.
Recall that Feature-based Tabu Search (Algorithm 15, in Section 2.5) operated by identifying
“features” in solutions found in good solutions, and then made those features “taboo”, temporarily
banned from being returned to by later Tweaks. The idea was to prevent the algorithm from revisit-
ing, over and over again, those local optima in which those features tended to be commonplace.
If you can construct a good, closed Tweak operator, it turns out that Feature-based Tabu Search
can be nicely adapted to the combinatorial optimization problem. Simply define “features” to be
the components of the problem. For example, Feature-based Tabu Search might hill-climb through
the space of Traveling Salesman Problem solutions, temporarily making certain high-performing
edges taboo to force it out of local optima in the TSP.
A variant of Feature-based Tabu Search called Guided Local Search (GLS) seems to be particu-
larly apropos for combinatorial optimization: it assigns “historical quality” measures to compo-
nents, like Ant Colony Optimization does. But interestingly, it uses this quality information not to
home in on the best components to use, but rather to make troublesome components taboo and
force more exploration.
GLS is by Chris Voudouris and Edward Tsang.126 The algorithm is basically a variation of Hill-
Climbing that tries to identify components which appear too often in local optima, and penalizes
later solutions which use those components so as to force exploration elsewhere.
125 Thishas been noted before, and not just by me: see p. 57 of Marco Dorigo and Thomas Stützle, 2004, Ant Colony
Optimization, MIT Press. So we’ve got similarities to coevolution and to EDAs... hmmm....
126 Among the earlier appearances of the algorithm is Chris Voudouris and Edward Tsang, 1995, Guided local search,
Technical Report CSM-247, Department of Computer Science, University of Essex. This technical report was later
updated as Chris Voudouris and Edward Tsang, 1999, Guided local search, European Journal of Operational Research, 113(2),
469–499.
152
To do this, Guided Local Search maintains a vector of pheromones,127 one per component, which
reflect how often each component has appeared in high-quality solutions. Instead of hill-climbing
by Quality, GLS hill-climbs by an AdjustedQuality function which takes both Quality and the presence
of these pheromones into account.128 Given a candidate solution S, a set of components C for the
problem, and a vector !p of current pheromones, one per component, the adjusted quality of S is
defined as:
(
pi if component Ci is found in S
AdjustedQuality(S, C, !p) = Quality(S) − β ∑
i 0 otherwise
Thus the hill-climber is looking for solutions both of high quality but also ones which are relatively
novel: they use components which haven’t been used much in high-quality solutions before. High
pheromones are bad in this context. The parameter β determines the degree to which novelty figures
in the final quality computation, and it will need to be tuned carefully.
After doing some hill-climbing in this adjusted quality space, the algorithm then takes its
current candidate solution S, which is presumably at or near a local optimum, and increases the
pheromones on certain components which can be found in this solution. To be likely to have its
pheromones increased, a component must have three qualities. First, it must appear in the current
solution — that is, it’s partly responsible for the local optimum and should be avoided. Second,
it will tend to have lower value or higher cost: we wish to move away from the least important
components in the solution first. Third, it will tend to have lower pheromones. This is because GLS
doesn’t just want to penalize the same components forever: it’d like to turn its attention to other
components for some exploration. Thus when a component’s pheromone has increased sufficiently,
it’s not chosen for further increases. Spread the love!
To determine the components whose pheromones should be increased, GLS first computes the
penalizability of each component Ci with current pheromone pi as follows:129
1
Penalizability(Ci , pi ) =
(1 + pi ) × Value(Ci )
...or if your problem is using costs...
Cost(Ci )
Penalizability(Ci , pi ) =
(1 + p i )
Guided Local Search then picks the most penalizable component presently found in the current
solution S and increments its pheromone pi by 1. If there’s more than one such component (they’re
tied), their pheromones are all increased.
Compare the Penalizability function with the Desirability function in Section 8.3.1: note that
components with high Desirability generally have low Penalizability and vice versa. While ACO
seeks to build new candidate solutions from historically desirable components, GLS punishes
components which have often appeared in local optima, though the ones it punishes the most are
the least desirable such components.
127 I’m borrowing ACO terminology here: GLS calls them penalties.
128 In the name of consistency I’m beginning to deviate from the standard GLS formulation: the algorithm traditionally
is applied to minimization rather than maximization problems.
129 GLS traditionally uses the term utility rather than my made-up word penalizability. Utility is a highly loaded term
that usually means something quite different — see Section 10 for example — so I’m avoiding it.
153
Now that we have a way to adjust the quality of solutions based on pheromones, and a way to
increase pheromones for components commonly found in local optima, the full algorithm is quite
straightforward: it’s just hill-climbing with an additional, occasional, adjustment of the current
pheromones of the components. There’s no evaporation (which is quite surprising!).
Guided Local Search doesn’t specify how we determine that we’re stuck in a local optimum
and must adjust pheromones to get ourselves out. Usually there’s no test for local optimality.
Thus below the approach I’ve taken is borrowed from Algorithm 10 (Hill-Climbing with Random
Restarts, Section 2.2), where we hill-climb until a random timer goes off, then update pheromones
under the presumption that we’ve hill-climbed long enough to roughly get ourselves trapped in a
local optimum.
Algorithm 113 Guided Local Search (GLS) with Random Updates

1: C ← {C1 , ..., Ci } set of possible components a candidate solution could have
3: !p ← $ p1 , ..., pi % pheromones of the components, initially zero

5: Best ← S
6: repeat
8: repeat " First do some hill-climbing in the pheromone-adjusted quality space
10: if Quality(R) > Quality(Best) then
11: Best ← R
12: if AdjustedQuality(R, C, !p) > AdjustedQuality(S, C, !p) then
13: S←R
14: until Best is the ideal solution, time is up, or we have run out of time
15: C ) ← {}
16: for each component Ci ∈ C appearing in S do " Find the most penalizable components
17: if for all Cj ∈ C appearing in S, Penalizability(Ci , pi ) ≥ Penalizability(Cj , p j ) then
18: C ) ← C ) ∪ {Ci }
19: for each component Ci ∈ C appearing in S do " Penalize them by increasing their pheromones
20: if Ci ∈ C ) then
21: pi ← pi + 1
23: return Best
The general idea behind Guided Local Search doesn’t have to be restricted to hill-climbing: it
could be used for population-based methods as well (and indeed is, where one version is known as
the Guided Genetic Algorithm).
154
9 Optimization by Model Fitting
Most of the methods we’ve examined so far sample the space of candidate solutions and select the
high-quality ones. Based on the samples, new samples are generated through Tweaking. Eventually
the samples (if we’re lucky) start migrating towards the fitter areas in the space.
But there’s an alternative to using selection and Tweak. Instead, from our samples we might
build a model (or update an existing one) which gives us an idea of where the good areas of the
space are. From that model we could then generate a new set of samples.
Models can take many forms. They could be neural networks or decision trees describing how
good certain regions of the space are. They could be sets of rules delineating regions in the space.
They could be distributions over the space suggesting where most of the population should go.
The process of fitting a model (sometimes known as a hypothesis) to a sample of data is commonly
known as induction, and is one of the primary tasks of machine learning.
This model building and sample generation is really just an elaborate way of doing selection and
Tweaking, only we’re not generating children directly from other individuals, but instead created
uniformly from the region in which the fitter individuals generally reside.
Much of the model-fitting literature in the metaheuristics community has focused on models
in the form of distributions, especially simplified distributions known as marginal distributions.
This literature is collectively known as Estimation of Distribution Algorithms (EDAs). But there
are other approaches, largely cribbed from the machine learning community. We’ll begin with one
such alternative, then get to EDAs afterwards.
9.1 Model Fitting by Classification

A straightforward way to fit a model to a population is to simply
divide the population into the fit individuals and the unfit individuals,
then tell a learning method to use this information to identify the
“fitter” regions of the space as opposed to the “unfit” regions. This is
basically a binary classification problem.130
One of the better-known variations on model-fitting by classifi-
cation is the Learnable Evolution Model (LEM), by Ryszard Michal-
ski.131 The overall technique is very simple: first, do some evolution.
Then when your population has run out of steam, break it into two Figure 55 Model fitting by clas-
groups: the “fit” and “unfit” individuals (and possibly a third group of sification via a decision tree.
“middling” individuals). Use a classification algorithm to identify the Black circles are “fit” and white
circles are “unfit” individuals
regions of the space containing the fit individuals but not containing in the population. The learned
the unfit individuals. Replace the “unfit” individuals with individuals model delimits fit and unfit re-
sampled at random from those identified regions. Then go back to do gions of the genotype space.
some more evolution.
There are plenty of binary classification algorithms available in the machine learning world: for
example, decision trees, Support Vector Machines (SVMs), k-Nearest-Neighbor (KNN), even
130 Classification is the task of identifying the regions of space which belong to various classes (or categories). Here, we
happen to be dividing the genotype space into two classes: the fit individuals class and the unfit individuals class. Hence
the term binary classification.
131 Ryszard Michalski, 2000, Learnable evolution model: Evolutionary processes guided by machine learning, Machine
Learning, 38(1–2), 9–40.
155
Michalski’s own AQ132 algorithm. LEM doesn’t care all that much. Figure 55 shows the results
of applying a decision tree to divide up the “fit” from “unfit” regions. Note some portions of the
space could have been fit better: part of this is due to the particular learning bias of the decision
tree algorithm, which emphasizes rectangles. Every learning method has a bias: pick your poison.
The algorithm:
Algorithm 114 An Abstract Version of the Learnable Evolution Model (LEM)

1: b ← number of “best individuals”
2: w ← number of “worst individuals” " b + w ≤ || P||. If you wish, you can make b + w = || P||.

4: Best ← !
5: repeat
6: repeat " Do some evolution
7: AssessFitness(P)
10: Best ← Pi
12: until neither P nor Best seem to be improving by much any more
13: P+ ⊂ P ← fittest b individuals in P " Fit a model
14: −
P ⊂ P ← least fit w individuals in P
15: M ← learned model which describes the space containing members of P+ but not P−
16: Q ←w children generated randomly from the space described in M " Generate children
17: P ← Join(P, Q) " Often P ← ( P − P− ) ∪ Q
19: return Best
Some notes. First, the Join operation in Line 17 is often done by simply replacing the w worst
individuals in P, that is, P− , with the Q new children. In other words, P ← ( P − P− ) ∪ Q. But you
could do Join in other ways as well. Second, M could also be based not on P but on all previously
tested individuals: why waste information?
Third, it’s plausible, and in fact common, to do no evolution at all, and do only model building:
that is, eliminate Line 11 (and do the inner loop only once). This model-building-only approach
will be used in later algorithms in this Section. Or, since it’s sometimes hard to determine if
things are “improving”, you could jut run the evolution step for some n times and then head into
model-building, or apply a timer a-la Hill-Climbing with Random Restarts (Algorithm 10).
Generating Children from the Model The models produced by classification algorithms fall
into two common categories: generative models and discriminative models. Generative models
can easily generate random children for you. Discriminative models cannot. But many common
classification algorithms (including all mentioned so far) produce discriminative models! What to
do? We could apply rejection sampling to our discriminative models: repeatedly generate random
individuals until one falls in the “high fitness” region according to our model.
132 Originally called Aq , later restyled as AQ. I don’t know why.
156
Algorithm 115 Simple Rejection Sampling
1: n ← desired number of samples
2: M ← learned model
3: P ← {}
4: for n times do
5: repeat
6: S ← individual generated uniformly at random
7: until S is in a “fit” region as defined by M
8: P ← P ∪ {S}
9: return P
As the run progresses and the population homes in on the optima in the space, the regions of
“fit” individuals become very small, and rejection sampling starts getting expensive. Alternatively,
you could try to gather the list of regions that are considered valid, and sample from them according
to their size. Imagine that you’ve gone through the model (a decision tree say) and have gathered a
list of “fit” regions. For each region you have computed a volume. You could perform a kind of
region-based sampling where you first pick a region proportional to their volumes (using Fitness
Proportionate Selection, but with volumes rather than fitnesses), and then select a point uniformly
at random within the chosen region. This would also create an entirely uniform selection.
Algorithm 116 Region-based Sampling
3: P ← {}
4: R ← { R1 , ..., Rm } “fit” regions from M, each with computed volumes
5: for n times do
6: Ri ← selected from R using Volume-Proportionate Selection " (Like algorithm 30, so to speak)
7: P ← P ∪ { individual generated uniformly from within the bounds of Ri }
8: return P
It turns out that many discriminative models don’t just create boundaries delimiting regions,
but really define fuzzy functions specifying the probability that a given point belongs to one class
or another. Deep in the “low fitness” regions, the probability of a point being “high fitness” is
very small; while deep in the “high fitness” regions it’s quite big. On the borders, it’s half/half.
Furthermore, there exist approximate probability estimation functions even for those algorithms
which are notionally boundary-oriented, such as k-Nearest-Neighbor, SVMs, and decision trees. For
example, in a decision tree, the probability of a region belonging to the “high fitness” class could be
assumed to be proportional to the number of “high fitness” individuals from the population from
which we built the model which were located in that region.
Assuming we have this probability, we could apply a weighted rejection sampling, where we
keep kids only with a probability matching the model:
157
(a) A population of 5 indi- (b) A population of 20 indi- (c) A population of 75 indi- (d) A distribution of an infi-
viduals viduals vdiuals nite number of individuals,
with Subfigure (c) overlaid
for reference.
Figure 56 The distribution of a population candidate solutions, using samples of 5, 20, and 75, plus an infinite population
distribution.
Algorithm 117 Weighted Rejection Sampling

3: P ← {}
4: for n times do
5: repeat
6: S ← individual generated uniformly at random
7: p ← probability that S is “fit”, according to M
8: until p ≥ random number chosen uniformly from 0.0 to 1.0 inclusive
9: P ← P ∪ {S}
10: return P
Algorithm 115 (simple rejection sampling) is just a degenerate version of weighted sampling,
where the probability is 1.0 if you’re in the “fit” region and 0.0 of you’re in the “unfit” region.
9.2 Model Fitting with a Distribution

An alternative form of model is a distribution of an infinite-sized population using some math-
ematical function. This is the basic idea behind Estimation of Distribution Algorithms (EDAs).
To conceptualize this, let’s begin with an approach to a distribution which in fact no EDAs (to
my knowledge) use, but which is helpful for illustration. Figure 56(a) shows a population of 5
individuals sampling the space roughly in proportion to the fitness of those regions. Figure 56(b)
has increased this to 20 individuals, and Figure 56(c) to 75 individuals. Now imagine that we
keep increasing the population clear to ∞ individuals. At this point our population has become an
infinite distribution of the sort shown in Figure 56(d), with different densities in the space. Thus in
some sense we may view Figures 56(a), (b), and (c) as sample distributions of the true underlying
infinite distribution shown in Figure 56(d).
158
That’s basically what a population actually is: in an ideal world we’d have an infinite number
of individuals to work with. But we can’t, because, well, our computers can’t hold that many. So
we work with a sample distribution instead.
The idea behind an Estimation of Distribution Algorithm is to represent that infinite population
in some way other than with a large number of samples. From this distribution we will typically
sample a set of individuals, assess them, then adjust the distribution to reflect the new fitness results
we’ve discovered. This adjustment imagines that the entire distribution is undergoing selection133
such that fitter regions of the space increase in their proportion of the distribution, and the less
fit regions decrease in proportion. Thus the next time we sample from the distribution, we’ll be
sampling more individuals from the fitter areas of the space (hopefully).
Algorithm 118 An Abstract Estimation of Distribution Algorithm (EDA)

1: D ← Build Initial Infinite Population Distribution
2: Best ← !
3: repeat
4: P ← a sample of individuals generated from D
5: AssessFitness(P)
8: Best ← Pi
9: D ← UpdateDistribution(D, P) " Using P’s fitness results, D undergoes “selection”
11: return Best
At this point you may have noticed that estimation of distribution algorithms are really just a
fancy way of fitting generative models to your data. Such models are often essentially telling you
the probability that a given point in space is going to be “highly fit”. Because they’re generative,
we don’t need to do rejection sampling etc.: we can just produce random values under the models.
In theory.
Representing Distributions So far we’ve assumed our space is real-

valued and multidimensional. Let’s go with that for a while. How
could you represent an infinite distribution over such a monster? One
way is to represent the distribution as an n-dimensional histogram.
That is, we discretize the space into a grid, and for each grid point we
indicate the proportion of the population which resides at that grid
point. This approach is shown in Figure 57. The difficulty with this
method is twofold. First, we may need a fairly high-resolution grid
to accurately represent the distribution (though we could do better by
allowing the grid squares to vary in size, as in a kd-tree or quadtree).
Second, if we have a high dimensional space, we’re going to need a Figure 57 Approximating the
lot of grid points. Specifically, if we have n genes in our genome, and distribution in Figure 56(d) with
each has been discretized into a pieces, we’ll need an numbers. Eesh. a histogram.
133 As it’s an infinite population, Tweaking is not actually necessary. Just selection.
159
Another way to represent our infinite population is with some kind
of parametric distribution. For example, we could use some m number
of gaussian curves to approximate the real distribution as shown in
Figure 58 (with m = 3). This has the advantage of not requiring a
massive number of grid squares. But it too has some problems. First
off, how many gaussian curves do we need to accurately describe this
population? Second, gaussian curves may not give you the cost savings
you were expecting. A one-dimensional gaussian, like everyone’s seen
in grade school, just needs a mean µ and variance σ2 to define it.
But in an n-dimensional space, a multinomial gaussian which can be
stretched and tilted in any dimension requires a mean vector !µ of size Figure 58 Approximating the
n and a covariance matrix134 Σ which is n2 in size. So if you have distribution in Figure 56(d)
1000 genes, you need a covariance matrix of size 1,000,000 for a single with three multivariate Gaus-
sian curves.
gaussian.
Still though, n2 is lots better than an . But it’s not nearly good enough. Thus most estimation of
distribution algorithms cheat and use a different representation which is radically simpler but at a
huge cost: a set of marginal distributions.
A marginal distribution is a projection of the full distribution onto (usually) a single dimension.
For example, Figure 59 shows the projection of the full joint distribution in two different directions,
one for x and one for y. If we just use the marginal distributions in each dimension, then instead of
a joint distribution of n dimensions, we just have n 1-dimensional distributions. Thus a marginal
distribution contains proportions of an infinite population which contain the various possible
values for a single gene. There is one marginal distribution per gene.
We’ve not come up with a new representation: just a
way to reduce the dimensionality of the space. So we’ll
still need to have some way of representing each of the
Marginal Distribution of y Gene

marginal distributions. As usual, we could use (for ex-
ample) a parametric representation like one or more 1-
dimensional gaussians; or we could use a 1-dimensional
array as a histogram, as shown in Figure 60.
From Figure 59 it appears that we could probably get
away with representing each marginal distribution with,
it appears, roughly two 1-dimensional gaussians. Each Joint Distribution
such gaussian requires a mean and a variance: that’s just of Vector Genotype〈x, y〉
8 numbers (a mean and variance for each gaussian, two
gaussians per marginal distribution, two marginal distribu-
tions). In general, if we needed b gaussians per dimension, Marginal Distribution of x Gene
we’d need 2bn numbers. A tiny amount compared to n2 .
Or if we chose to use a histogram, discretizing our one- Figure 59 Marginalized versions of the distri-
bution in Figure 56(d). Since each distribution
dimensional distributions each into b buckets, that’s still has two peaks, each could probably be rea-
bn numbers, instead of the an we needed for the joint his- sonably approximated with two gaussians per
togram. Great! (Actually, there’s an ugly problem, but we’ll distribution (four in total).
get to that in a bit.)
134 Yes.
Σ is classically used to represent covariance matrices. Not to be confused with the summation symbol ∑. Ugh.
Try summing covariance matrices some time: ∑i ∑ j Σij . Wonderful.
160
Now that we’ve burned out on real-valued spaces, consider (finite135 ) discrete spaces. Rep-
resenting a joint discrete space is exactly like the grid in Figure 57, except (of course) don’t need
to discretize: we’re already discrete. However we still have a potentially huge number of points,
making attractive the marginal distributions again. Each marginal distribution is, as usual, a
description of the fractions of the population which have a particular value for their gene. Each
gene thus has a marginal distribution consisting of just an array of fractions, one for every possible
gene value. Similar to the marginalized histogram example.
In fact, if you have w possible gene values, you don’t really need an array of size w. You just
need the first w − 1 elements. The array must sum to 1 (it’s a distribution), so it’s clear what the last
element value is.
We can get even simpler still: what if our space is sim-
ply multidimensional boolean? That is, each point in space
is just a vector of booleans? You couldn’t get simpler: the
marginal distribution for each gene is just a single num-
ber: the fraction of the population which has a 1 in that
gene position (as opposed to a 0). Thus you can think of
all marginal distributions for an n dimensional boolean
problem as a single real-valued vector of length n, with
Figure 60 Gaussian and histogram represen-
each value between 0.0 and 1.0. tations of a 1-dimensional marginal distribu-
tion.
9.2.1 Univariate Estimation of Distribution Algorithms
Now that we have a way of reducing the space complexity through marginalization, and can
represent marginal distributions in various ways, we can look at some actual EDAs. The first EDAs
were univariate EDAs: they used the marginalizing trick described earlier. Most of them also
operated over discrete or even boolean spaces.
Among the earliest such EDAs was Population-Based Incremental Learning (PBIL), by
Shumeet Baluja.136 PBIL assumes a finite discrete space. This algorithm begins with n marginal
distributions, one per gene. Each distribution is initially uniform, but that’ll change soon. The
algorithm then repeatedly samples individuals by picking one gene from each distribution. It
then assesses the fitness of the individuals, and applies truncation selection to throw out the worst
ones. It then updates each marginal distribution by throwing out a little of its old probabilities
and rolling in a little of the proportions of values for that gene which exist among the remaining
(fitter) individuals. We then throw away the individuals and go back to making new ones from the
revised distribution.
135 Countably infinite spaces, like the space of all integers or the space of trees or graphs, present a much yuckier
problem and typically aren’t handled by EDAs.
136 The first PBIL document was Shumeet Baluja, 1994, Population-based incremental learning: A method for integrating
genetic search based function optimization and competitive learning, Technical Report CMU-CS-94-163, Carnegie Mellon
University. The first formal publication, with Rich Caruana, was Shumeet Baluja and Rich Caruana, 1995, Removing the
genetics from the standard genetic algorithm, in Armand Prieditis and Stuart Russell, editors, Proceedings of the Twelfth
International Conference on Machine Learning (ICML), pages 38–46, Morgan Kaufmann.
161
Algorithm 119 Population-Based Incremental Learning (PBIL)
1: popsize ← number of individuals to generate each time
2: b ← how many individuals to select out of the generated group
3: α ← learning rate: how rapidly to update the distribution based on new sample information
4: D ← { D1 , ..., Dn } marginal distributions, one per gene " Each uniformly distributed
5: Best ← !
6: repeat
7: P ← {} " Sample from D
9: S ← individual built by choosing the value for each gene j at random under distribution D j
12: Best ← S
13: P ← P ∪ {S}
14: P ← the fittest b individuals in P " Truncation selection
15: for each gene j do " Update D
16: N ← distribution over the possible values for gene j found among the individuals in P
17: D j ← (1 − α) D j + αNj
19: return Best
That last equation ( D j ← (1 − α) D j + αNj ) deserves some explanation. Keep in mind that
because PBIL operates over discrete spaces, each distribution D j is just a vector of fractions, one for
each value that gene j can be. We multiply each of these fractions by 1 − α, and add in α’s worth of
fractions from Nj . Nj is the vector, one per value that gene j can be, of the fraction of members of P
that have that particular value for gene j. So α helps us to gradually change the distribution.
In short: we sampled from D, threw out the least fit samples, and rolled their resulting distribu-
tions back into D. As a result D has shifted to be closer to the fitter parts of the space.
A variation on PBIL is the Univariate Marginal Distribution Algorithm (UMDA), by Heinz
Mühlenbein.137 UMDA differs from PBIL only in two small respects. First, UMDA doesn’t specify
truncation selection as the way to reduce P: any selection procedure is allowed. Second, UMDA
entirely replaces the distribution D each time around. That is, α = 1. Because there’s no gradualness,
if our latest sample doesn’t contain a given value for a certain gene, that value is lost forever, just
like using crossover without mutation in the genetic algorithm. As a result, to maintain diversity
we will require a large sample each time if the number of discrete values each gene can take on is
large. Perhaps for this reason, UMDA is most often applied to boolean spaces.
Next, we consider the Compact Genetic Algorithm (cGA) by Georges Harik, Fernando Lobo,
and David Goldberg, which operates solely over boolean spaces.138 cGA is different from PBIL
in important ways. Once again, we have a distribution and use it to generate some collection
of individuals, but rather than do selection on those individuals, we instead compare every pair
137 Heinz Mühlenbein, 1997, The equation for response to selection and its use for prediction, Evolutionary Computation,
5(3), 303–346.
138 It’s never been clear to me why it’s cGA and not CGA. Georges Harik, Fernando Lobo, and David Goldberg, 1999,
The compact genetic algorithm, IEEE Transactions on Evolutionary Computation, 3(4), 287–297.
162
of individuals Pi and Pk in our sample. Assume Pi is fitter. For each gene j, if Pi and Pk differ in
value at gene j, we shift D j so that it will generate Pi ’s gene value more often in the future. Since
cGA works only with booleans, gene values can only be 1 or 0, and each distribution D j is just a
real-valued number (how often we pick a 1 versus a 0). If Pi was 1 and Pk was 0, we increase D j by
a small amount. Thus not only do the “fit” individuals have a say in how the distribution changes,
but the “unfit” individuals do as well: they’re telling the distribution: “don’t be like me!”
The cGA doesn’t model an infinite population, but rather a very large but finite population.
1
Thus the cGA has “steps” for incrementing or decrementing distributions, each step discretization in
size. Moving one step up in a discretization represents one more member of that large population
taking on that particular gene value. Though I’m not sure why you couldn’t just say
D j ← (1 − α) D j + α(value of gene j in Pi − value of gene j in Pk )
(or in the notation of the algorithm below, use U and V instead of Pi and Pk ).
Algorithm 120 The Compact Genetic Algorithm (cGA)
1: popsize ← number of individuals to generate each time
2: discretization ← number of discrete values our distributions can take on " Should be odd, ≥ 3
3: D ← { D1 , ..., Dn } marginal boolean distributions, one per gene " Each uniform: set to 0.5
4: gameover ← false
5: Best ← !
6: repeat
7: if for all genes j, D j = 1 or D j = 0 then " D has converged, so let’s quit after this loop
8: gameover ← true
9: P ← {} " Sample from D
11: S ← individual built by choosing the value for each gene j at random under distribution D j
14: Best ← S
15: P ← P ∪ {S}
16: for i from 1 to || P|| do " For all pairs Pi and Pk , i 4= k...
17: for k from i + 1 to || P|| do
18: U ← Pi
19: V ← Pk
20: if Fitness(V) > Fitness(U) then " Make sure U is the fitter individual of the two
21: Swap U and V
22: for each gene j do " Update each D j only if U and V are different
23: if the value of gene j in U > the value of gene j and D j < 1 then " 1 vs. 0
1
24: D j ← D j + discretization " Push closer to a 1
25: else if the value of gene j in U < the value of gene j and D j > 1 then " 0 vs. 1
1
26: D j ← D j − discretization " Push closer to a 0
27: until Best is the ideal solution, or gameover=true, or we have run out of time
28: return Best
163
I augmented this with our standard “Best” mechanism: though in fact the cGA doesn’t normally
include that gizmo. Instead the cGA normally runs until its distributions are all 1’s or 0’s, which
indicates that the entire “population” has converged to a given point in the space. Then it just
returns that point (this is easily done by just sampling from the D j distributions one last time). To
augment with the “Best” mechanism, I’m just running the loop one final time (using the gameover
counter) to give this final sampling a chance to compete for the “Best” slot.
The version of cGA shown here is the more general “round robin tournament” version, in
which every individual is compared against every other individual. A more common version of
cGA just generates two individuals at a time and compares them. This can be implemented simply
by setting the size of P to 2 in the round-robin tournament version.
In the round robin tournament version, we have to ensure that 0 ≤ D j ≤ 1; but in the || P|| = 2
version, it so happens that happens automagically. When D j reaches (say) 0, then 100% of the
individuals sampled from it will have 0 in that gene slot. That includes U and V. U and V will now
always have the same value in that slot and the if-statements (lines 22 and 24) will be turned off.
Real-Valued Representations So far we’ve seen algorithms for boolean and discrete marginal
distributions. How about real-valued ones?
Once we’ve marginalized a real-valued distribution, we’re left with m separate 1-dimensional
real-valued distributions. As discussed earlier, we could just discretize those distributions, so each
gene would have some m (discrete) gene values. At this point we could just use PBIL: generate an
individual by, for each gene, first picking one of those discrete gene values, then picking a random
real-valued number within that discretized region. Likewise, to determine if a (discretized) gene
value is found in a given individual, you just discretize the current value and see if it matches.
There are other approaches too. For example, you could represent each marginal distribution
with a single gaussian. This would require two numbers, the mean µ and variance σ2 , per distribu-
tion. To create an individual, for each gene you just pick a random number under the gaussian
distribution defined by µ and σ2 , that is, the Normal distribution N (µ, σ2 ) (see Algorithm 12).
In PBIL, to adjust the distribution to new values of µ and σ2 based on the fitness results, we
first need to determine the mean µ Nj and variance σN 2 of the distribution N described by the fit
j j
individuals stored in P. The mean is obvious:
1
|| P|| P∑
µ Nj = value of gene j of Pi
∈P i
We could use the unbiased estimator139 for our variance:

1
2
σN j
=
|| P|| − 1 ∑ (value of gene j of Pi − µ Nj
)2
Pi ∈ P
Now we just update the distribution D j . Instead of using this line:

D j ← (1 − α) D j + αNj
We could do:
µ Dj ← (1 − α)µ Dj + αµ Nj
2
σD j
← (1 − β)σD2 j + βσN
2
j
139 I think this is what we want. If it isn’t, then it’s || P1 || rather than || P||−
1
1
.
164
The idea is to make the distribution in D j more similar to the sample distribution we gathered
in Nj . To be maximally general, σ2 has its own learning rate β, but if you like you could set β = α.
Of course, in Figure 59 the distributions weren’t described easily with a single gaussian, but
rather would be okay with two gaussians each. Updating a multimodal distribution like that
is perfectly doable but trickier, involving a variant of gradient descent called the Expectation
Maximization or EM algorithm. That’s a whole topic in and of itself, so I’ll just leave it there. But
in truth, I’d use several gaussians per marginal distribution in most cases.
9.2.2 Multivariate Estimation of Distribution Algorithms

There is a very big problem with using marginal distributions, and it turns out it is the exact
same problem that is faced by Cooperative Coevolution: it assumes that there is no linkage at all
between genes. Each gene can be relegated to its own separate distribution without considering
the joint distribution between the genes. We’re throwing information away. As a result, marginal
distributions suffer from essentially the same maladies that Cooperative Coevolution does.140 As a
result, univariate EDAs may easily get sucked into local optima for many nontrivial problems.
Recognizing this problem, recent EDA research has focused on coming up with more sophis-
ticated EDAs which don’t just use simple marginal distributions. But we can’t just go to the full
joint distribution: it’s too huge. Instead they’ve moved a little towards the joint by using bivariate
distributions: one distribution for every pair of genes in the individual. If you have n genes,
this results in n2 − n distributions, and that’s prohibitively expensive. And if we go to triples or
quadruples of genes per distribution, it gets uglier still.
Various algorithms have been proposed to deal with this. The prevailing approach seems to
be to find the pairs (or triples, or quadruples, etc.) of genes which appear to have the strongest
linkage, rather than computing all the combinations. We can then represent the joint distribution
of the space as a collection of univariate distributions and the most strongly-linked bivariate (or
multivariate) distributions. This sparse approximate representation of the space is known as a
Bayes Network. Now instead of building the distribution D from our samples, we build a Bayes
Network N which approximates the true distribution D as well as possible. N is likewise used to
generate our new collection of samples.
Here I’m going to disappoint you: I’m not going explain how to build a Bayes Network from a
collection of data (in our case, a small population), nor explain how to generate a new data point
(individual) from the same. There is an entire research field devoted to these topics. It’s complex!
And depending on the kind of data (real-valued, etc.), and the models used to represent them
(gaussians, histograms, whatnot), it can get much more complex still. Instead, it might be wise to
rely on an existing Bayes Network or Graphical Model package to do the hard work for you.
With such a package in hand, the procedure is pretty easy. We begin with a random sample
(population) and cut it down to just the fitter samples. We then build a network from those samples
140 The model behind Cooperative Coevolution is basically identical to univariate estimation of distribution algorithms
in its use of marginalization. The only difference is that Cooperative Coevolution uses samples (individuals in popula-
tions) for its “marginal distributions”, while univariate EDAs use something else — gaussians, histograms, what have
you. Compare Figures 44 and 45 in the Coevolution Section with Figure 59 showing marginalized distributions: they’re
very similar. Christopher Vo, Liviu Panait, and I had a recent paper on all this: Christopher Vo, Liviu Panait, and Sean
Luke, 2009, Cooperative coevolution and univariate estimation of distribution algorithms, in FOGA ’09: Proceedings of the
Tenth ACM SIGEVO Workshop on the Foundations of Genetic Algorithms, pages 141–150, ACM. It’s not a giant result but it
was fun to write.
165
which approximates their distribution in the space. From this distribution we generate a bunch
of new data points (the “children”). Then the children get joined into the population. This is the
essence of the Bayesian Optimization Algorithm (BOA) by Martin Pelikan, David Goldberg, and
Eric Cantú-Paz. A more recent version, called the Hierarchical Bayesian Optimization Algorithm
(hBOA),141 is presently the current cutting edge, but BOA suffices for our purposes here:
Algorithm 121 An Abstract Version of the Bayesian Optimization Algorithm (BOA)

1: p ← desired initial population size
2: µ ← desired parent subset size
3: λ ← desired child subset size
4: Best ← !
5: P ← { P1 , ..., Pp } Build Initial Random Population
6: AssessFitness(P)
9: Best ← Pi
10: repeat
11: Q ⊆ P ← Select µ fit individuals from P " Truncation selection is fine
12: N ← construct a Bayesian Network distribution from Q
13: R ← {}
14: for λ times do
15: R ← R ∪ { individual generated at random under N }
16: AssessFitness(R)
17: for each individual R j ∈ R do
18: if Fitness(R j ) > Fitness(Best) then
19: Best ← R j
20: P ← Join(P, R) " You could do P ← Q ∪ R, for example
22: return Best
So what’s really going on with algorithms like these? They’re actually little more than ex-
travagant methods for doing population resampling. But they’re different in an important way:
the Bayes Network is essentially finding not just highly fit individuals to resample into a new
population, it’s trying to identify why they’re highly fit. What features do they appear to have in
common? Which elements in the individuals appear to matter and which ones don’t?
This is a big deal: it can home in on the best parts of the space fairly rapidly. But it comes at a
considerable cost: Algorithms along these lines can get very complex due to manipulation of the
Bayes Network, particularly if the space isn’t something simple like a boolean space.
141 Likewise, I have no idea why it’s not HBOA. The BOA algorithm was introduced in Martin Pelikan, David E.
Goldberg, and Erick Cantú-Paz, 1999, BOA: The bayesian optimization algorithm, in Wolfgang Banzhaf, et al., editors,
Proceedings of the Genetic and Evolutionary Computation Conference GECCO-1999, pages 525–532, Morgan Kaufmann. Two
years later, hBOA was published in Martin Pelikan and David E. Goldberg, 2001, Escaping hierarchical traps with
competent genetic algorithms, in Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2001), pages
511–518, Morgan Kaufmann.
166
10 Policy Optimization
Section 4.5.1 introduced the notion of an agent which follows a simple program called a policy.
Much of this section concerns methods for an agent to learn or optimize its policy.142 To do so,
the agent will wander about doing what an agent does, and occasionally receive a reward (or
reinforcement) to encourage or discourage the agent from doing various things. This reward
ultimately trickles back through earlier actions the agent did, eventually teaching the agent which
actions help to lead to good rewards and away from bad ones.
In the machine learning community, non-metaheuristic methods for learning policies are
well established in a subfield called reinforcement learning. But those methods learn custom
rules for every single state of the world. Instead, there are evolutionary techniques which do
this, known as Michigan-Approach Learning Classifier Systems (LCS) or Pitt-Approach Rule
Systems, which find much smaller, sparse descriptions of the entire state space. We’ll begin
by examining reinforcement learning because it is so closely associated with the evolutionary
methods both historically and theoretically. Specifically, we’ll spend quite a few pages on a
non-metaheuristic reinforcement learning method called Q-Learning. Then we’ll move to the
evolutionary techniques.
I won’t kid you. This topic can be very challenging to understand. You’ve been warned.143
10.1 Reinforcement Learning: Dense Policy Optimization

We begin with a non-metaheuristic set of techniques for learning dense policies, collectively known
as reinforcement learning, partly to put the metaheuristic methods (in Section 10.2) in context,
and partly because it teaches some concepts we’ll need as we go on.
Reinforcement learning is a strange term. Generally speaking, it refers to any method that learns
or adapts based on receiving quality assessments (the rewards or punishments — the reinforcement).
Thus every single topic discussed up to this point could be considered reinforcement learning.
Unfortunately this very general term has been co-opted by a narrow sub-community interested
in learning policies consisting of sets of if→then rules. Recall that in Section 4.5 such rules were
called state-action rules, and collectively described what to do in all situations the agent might
find itself in. The reinforcement learner figures out what the optimal state-action ruleset is for a
given environment, based solely on reinforcement received when trying out various rulesets in the
environment.
What kinds of environments are we talking about? Here’s an example: a cockroach robot’s
world is divided into grid squares defined by GPS coordinates. When the robot tries to move from
grid square to grid square (say, going north, south, east, or west), sometimes it succeeds, but with a
certain probability it winds up in a different neighboring square by accident. Some grid squares
block the robot’s path in certain directions (perhaps there’s a wall). In some grid locations there are
yummy things to eat. In other places the robot gets an electric shock. The robot does not know which
squares provide the food or the shocks. It’s just trying to figure out, for each square in its world,
142 Unlike most other topics discussed so far, this is obviously a specific application to which metaheuristics may be
applied, rather than a general area. But it’s included here because this particular application has spawned unusual and
important metaheuristics special to it; and it’s a topic of some pretty broad impact. So we’re going with it.
143 If you want to go deeper into Q-Learning and related methods, a classic text on reinforcement learning is Richard
Sutton and Andrew Barto, 1998, Reinforcement Learning: an Introduction, MIT Press. This excellent book is available online
at http://www.cs.ualberta.ca/∼sutton/book/the-book.html for free.
167
what direction should he go so as to maximize the yummy food and
minimize the shocks over the robot’s lifetime.
At right is a possible robot cockroach world, where if the cock-
roach stumbles into one area it gets a yummy treat (+1), and if it -2
stumbles into another area it gets an electric shock (−2).
In this example, the cockroach robot is our agent. The grid squares
are the external states (or just states) the agent may find itself in. The
directions the cockroach tries to move are the actions available to the
agent; different states may have different actions (in this case, because +1
of the presence of walls). The yummy things to eat are positive re-
inforcement or positive reward, and the electric shocks are likewise
Figure 61 Robot cockroach
negative reinforcement, or punishment, or negative reward (so to world with rewards (all unla-
speak). The agent’s attempt to maximize positive reinforcement over beled states have zero reward).
its lifetime is also known as trying to maximize the agent’s utility144
(or value). The probability of winding up in a new state based on the current state and chosen
action is known as the transition model. Our agent usually doesn’t know the transition model, but
one exists.
The reason each if→then rule is called a state-action rule in this context is because the if side
indicates a possible external state, and the then side indicates what action to take when in that
state. The agent is trying to construct a set of such rules, one for each possible external state, which
collectively describe all the actions to take in the world. This collection of rules is known as a
policy, and it is traditionally145 denoted as a function π (s) which returns the action a to take when
in a given state s. Figure 62 shows a likely optimal policy for the cockroach world.
Let’s do another example. We want to learn how to play Tic-Tac-
Toe (as X) against a random opponent based entirely on wins and
losses. Each possible board situation where X is about to play may
be considered a state. For each such state, there are some number of
! " # "
moves X could make; these are available actions for the state. Then
our opponent plays a random move against us and we wind up in a
" " "
new state: the probability that playing a given action in a given state
will wind up in a given new state is the transition model. Doing actions " " " #
in certain states wind up punishing us or rewarding us because they
cause us to immediately win or lose. Those are our reinforcements.
X X +
! # #
For example, if X plays at the location + in the state O − X
O ; O Figure 62 An optimal policy for
then X receives a positive reinforcement because X wins the game. If the cockroach robot world.
X plays at the location −, X probably loses immediately and receives
negative reinforcement provided the opponent isn’t stupid146 (keep
in mind, the next state is after the opponent makes his move too). And if X plays at ; then X
doesn’t get any reinforcement immediately as the game must still continue (for a bit). Not getting
reinforcement is also a kind of reinforcement: it’s just a reinforcement of zero. Ultimately we’re
trying to learn a policy which tells us what to do in each board configuration.
144 Not to be confused with utility in Section 8.4.
145 Yes,using π as a function name is stupid.
146 Of course, to get in this situation in the first place, our random opponent wasn’t the sharpest knife in the drawer.
168
Here’s a third example, stolen from Minoru Asada’s147 work in robot soccer. A robot is trying
to learn to push a ball into a goal. The robot has a camera and has boiled down what it sees into the
following simple information: the ball is either not visible or it is in the left, right, or center of the
field of view. If the ball is visible, it’s also either small (far away), medium, or large (near). Likewise
the goal is either not visible, on the left, right, or center, and if visible it’s either small, medium, or
large. All told there are ten ball situations (not visible, left small, left medium, left large, center
small, center medium, center large, right small, right medium, right large) and likewise ten goal
situations. A state is a pair of goal and ball situations: so there are 100 states. The robot can move
forward, curve left, curve right, move backward, back up to the left, and back up to the right. So
there are 6 actions for each state. The robot receives a positive reward for getting the ball in the
goal and zero for everything else.
It’s not just robots and games: reinforcement learning is in wide use in everything from factory-
floor decision making to gambling to car engines deciding when and how to change fuel injection
to maximize efficiency to simulations of competing countries or businesses. It’s used a lot.
All these examples share certain common traits. First, we have a fixed number of states. Second,
each state has a fixed number of actions, though the number and makeup of actions may differ
from state to state. Third, we’re assuming that performing an action in a given state transfers
to other states with a fixed probability. That’s nonsense but it’s necessary nonsense to make
the problem tractable. Fourth, we’re also assuming that we receive rewards for doing certain
actions in certain states, and that these rewards are either deterministic or also occur with a fixed
probability on a per state/action basis. That’s also a somewhat ridiculous assumption but keeps
things tractable. And now the final nonsense assumption: the transition probabilities are based
entirely on our current state and action — earlier actions or states do not influence the probabilities
except through the fact that they helped us to land in our current state and action. That is, to figure
out what the best possible action is for a given state, we don’t need to have any memory of what
we did a while back. We just need a simple if→then describing what to do given the situation
we are in now. This last assumption is commonly known as an assumption of a Markovian148
environment. Very few real situations are Markovian: but this assumption truly makes the problem
tractable, so we try to make it whenever possible if it’s not totally crazy.
10.1.1 Q-Learning
Q-Learning is a popular reinforcement learning algorithm which is useful to understand before we
get to the evolutionary models. In Q-Learning, the agent maintains a current policy π (s) (the best
policy it’s figured out so far) and wanders about its environment following that policy. As it learns
that some actions aren’t very good, the agent updates and changes its policy. The goal is ultimately
to figure out the optimal (smartest possible) policy, that is, the policy which brings in the highest
expected rewards over the agent’s lifetime. The optimal policy is denoted with π ∗ (s).
The agent doesn’t actually store the policy: in fact the agent stores something more general than
that: a Q-table. A Q-table is a function Q(s, a) over every possible state s and action a that could
be performed in that state. The Q-table tells us how good it would be to be presently in s, and
147 Among lots of other things, Minoru Asada is the co-founder of the RoboCup robot soccer competition.
148 Andrey Andreyevich Markov was a Russian mathematician from 1856–1922, and was largely responsible for Markov
chains, which are lists of states s1 , s2 , ... the agent finds itself in as it performs various actions in a Markovian environment.
This field, a major area in probability theory, is a large part of what known as stochastic processes, not to be confused
with stochastic optimization.
169
Best Actions for Further Possible
Initial State Initial Action Possible Those States Resulting States
Resulting States and Other Actions And So on
Probabilities ...
that doing a while in s .)
will land you in P(..
...
various states s!, such as "*(s! (1)) P(...)
s! (1), s! (2), s! (3) etc. P(...) ...
other a!
) s! (1)
(1) | s,a other a! ...
P(s! P(...)
"*(s! (2)) P(...) ...
s a P(s! (2) | s,a) s! (2) P(...)
other a! ...
P(s! (3)
| s,a other a!
) s! (3) ...
P(...)
"*(s! (3)) P(...)
...
other a! P(..
.)
other a! ...
Figure 63 The Q-Learning state-action model. We are presently in some state s and decide to perform an action a. With
a certain probability P(s) |s, a), doing that action a while in s leads to a state s) (here there are three possible s) we could
land in, s)(1) , s)(2) , and s)(3) ). We presume that from then on out we make the smartest possible action π ∗ (s) ) for each state
s) , leading to still further states and and smartest possible actions for them, and so on. Note that in this model the first
action we do (a) may not be the smartest action for s.
then perform action a, and then follow the optimal policy from then on. Thus the Q-value tells us the
utility of doing action a when in s if we were a perfect agent (other than our initial choice of a). The
agent starts with crummy Q-tables with lots of incorrect information, and then tries to update them
until they approach the optimal Q-table, denoted Q∗ (s, a), where all the information is completely
accurate. For a given state s, we would expect the best action a for that state (that is, π ∗ (s)) to have
a higher Q∗ value than the other actions. Thus we can define π ∗ (s) = argmaxa Q∗ (s, a), meaning,
“the action a which makes Q∗ (s, a) the highest”).
The world is a Markovian world: when an agent performs an action a in a given state s, the
agent will then transition to another state s) with a certain transition probability P(s) |s, a). The
agent also receives a reward R(s, a) as a result. Figure 63 shows the Q-learning state-action model:
an agent performs some action a, leading to one of several possible states s) , and we’ll assume
(perhaps wrongly) that the agent will choose perfect actions from π ∗ thereafter.
In a perfect world, where we actually knew P(s) |s, a), there’s a magic equation which we can
use to compute Q∗ (s, a):
Q∗ (s, a) = R(s, a) + γ ∑ P(s) |s, a) max Q∗ (s) , a) ) (1)

)
s) a
This equation says: the Q∗ value of doing action a while in state s is equal to the expected sum
of all future rewards received thereafter. This is equal to the first reward received, followed by the
170
Time = 0 Time = 1 Time = 2 Time = 3 Time = n
All Q values Certain actions get very Q values leak into actions Q values leak into actions which Q values
are initially 0 good (or bad) Q values which lead to the possibility lead to the actions leaked into have converged
because of rewards of high-reward actions at Time=2
$!"
Each grid Within a given state, each circle indicates a The color of the circle (black is very negative,
square is a possible action ($!"#) in that state. Some white is very positive, gray is zero) indicates
state. # states don!t have certain actions available. the Q value for that action in that state.
Figure 64 Illustration of Q-Learning with a model in the cockroach robot world. By time=n, the best (lightest color)
action in each state corresponds to the optimal policy action in Figure 62.
sum, over all possible new states s) we might land in, of the likelihood that we land there, times the
Q∗ value of the smartest action a) we could perform at that point. It’s a recursive definition.149
This is hard to get your head around, so think of it this way. At different times t the agent finds
itself in various states st and performs various actions at . For doing this, the agent receives various
rewards along the way (which might be 0). Thus the agent’s total rewards are:
R(s0 , a0 ) + R(s1 , a1 ) + R(s2 , a2 ) + R(s3 , a3 ) + ...
Let’s assume these are all optimal actions, and to keep things simple, there’s no probability:
when you do an action in a given state, you will wind up in a specific new state. Then the Q∗ (s2 , a2 )
value at time 2 is equal to the sum total rewards from then on, that is, R(s2 , a2 ) + R(s3 , a3 ) + ....
Similarly the Q∗ (s1 , a1 ) value at time 1 is equal to R(s1 , a1 ) + R(s2 , a2 ) + R(s3 , a3 ) + .... Thus
Q∗ (s1 , a1 ) = R(s1 , a1 ) + Q∗ (s2 , a2 ). Similarly, Q∗ (s0 , a0 ) = R(s0 , a0 ) + Q∗ (s1 , a1 ). See the similarity
with Equation 1? That equation had the additional term ∑s) P(s) |s, a) maxa) Q∗ (s) , a) ), rather than
just Q∗ (s) , a) ). This is because of the transition probability P. The term tells us the weighted average
Q∗ received in the future.
So what’s the γ? This is a cut-down constant between 0 and 1. It makes future rewards worth
less than earlier rewards, and without it, the Q∗ values could be infinite (which isn’t good).
If we had the P(s) |s, a) distribution and R(s, a) function at our disposal, we could use this magic
equation to figure out Q∗ . It works like this:
149 TheQ∗ equation is derived from a famous simpler equation by Richard Bellman called the Bellman Equation.
That equation doesn’t have actions explicitly listed, but rather assumes that the agent is performing some (possibly
suboptimal) hard-coded policy π. The Bellman equation looks like this:
U (s) = R(s) + γ max ∑ P(s) |s, a)U (s) )

a
s)
The U (s) bit is the equivalent of Q∗ (s, a), but it assumes that the a we do is always π (s). By the way, it’s U for Utility,
just as it’s R for Reward or Reinforcement. Sometimes instead of U you’ll see V (for the synonymous Value). The probability
function isn’t usually denoted P(s) |s, a) — I wrote it that way to be consistent with probability theory — but is rather
usually written T (s, a, s) ). That is, T for Transition Probability. Hmmm, I wonder we if could use Q for Q-tility...
171
Algorithm 122 Q-Learning with a Model
1: R ( S, A ) ← reward function for doing a while in s, for all states s ∈ S and actions a ∈ A
2: P ( S) | S, A ) ← probability distribution that doing a while in s results in s) , for all s, s) ∈ S and a ∈ A
3: γ ← cut-down constant " 0 < γ < 1. 0.5 is fine.
4: Q∗ (S, A) ← table of utility values for all s ∈ S and a ∈ A, initially all zero
5: repeat
6: Q) (S, A) ← Q∗ (S, A) " Copy the whole table
7: for each state s do
8: for each action a performable in s do
9: Q∗ (s, a) ← R(s, a) + γ ∑s) P(s) |s, a) maxa) Q) (s) , a) )
10: until Q∗ (S, A) isn’t changing much any more
11: return Q∗ (S, A)
That is, we start with absurd notions of Q∗ , assume they’re correct, and slowly fold in rewards
until our Q∗ values don’t change any more. This notion is called bootstrapping, and it may seem
crazy but it’s perfectly doable because of a peculiarity of Q-learning made possible by Markovian
environments: the Q-learning world has no local optima. Just one big global optimum. Basically
this is an obsfucated way of doing hill-climbing.
Q-Learning as Reinforcement Learning The algorithm just discussed is an example of what is

known in engineering and operations research circles as dynamic programming. This isn’t to be
confused with the use of the same term in computer science.150 In computer science, dynamic
programming is an approach to solve certain kinds of problems faster because they can be broken
into subproblems which overlap. In engineering, dynamic programming usually refers to figuring
out policies for agents in Markovian environments where the transition probability P and reward
function R are known beforehand.
From an artificial intelligence perspective, if we have P and R, this isn’t a very interesting
algorithm. Instead, what we really want is an algorithm which discovers Q∗ without the help of
P or R, simply by wandering around in the environment and, essentially, experiencing P and R
first-hand. Such algorithms are often called model-free algorithms, and reinforcement learning is
distinguished from dynamic programming by its emphasis on model-free algorithms.
We can gather Q∗ without P or R by discovering interesting facts from the environment as
we wander about. R is easy: we just fold in the rewards as we receive them. P is more complex
to explain. We need to replace the ∑s) P(s) |s, a) portion of Equation 1. This portion added in the
various Q∗ (s) , a) ) according to how often they occur. Instead, now we’ll just add them in as we
wind up in various s) . Wander around enough and the distribution of these s) approaches P(s) |s, a).
So: we’ll build up an approximation of Q∗ , based on samples culled from the world, called Q.
The table is initially all zeros. As we’re wandering about, we perform various actions in states,
transitioning us to new states and triggering rewards. Let’s say we’re in state s and have decided to
perform action a. Performing this action transitioned us to state s) and incurred a reward r. We
then update our Q table as:
150 Actuallythere’s a historical relationship between the two: but it’s a long story. Suffice it to say, the engineering usage
predates the computer science usage by quite a bit.
172
Q(s, a) ← (1 − α) Q(s, a) + α(r + γ max Q(s) , a) )) (2)
) a
Notice that we’re throwing away a bit of what we know so far, using the 1 − α trick — we
saw this before in Ant Colony Optimization (in Section 8.3) and in Estimation of Distribution
Algorithms (in Section 9.2.1) — and roll in a bit of the new information we’ve learned. This new
information is set up in what should by now be a familiar fashion: the reward r plus the biggest Q
of the next state s) . Notice the relationship to Equation 1. The revised algorithm is then:
Algorithm 123 Model-Free Q-Learning

1: α ← learning rate " 0 < α < 1. Make it small.
3: Q(S, A) ← table of utility values for all s ∈ S and a ∈ A, initially all zero
4: repeat
5: Start the agent at an initial state s ← s0 " It’s best if s0 isn’t the same each time.
6: repeat
7: Watch the agent make action a, transition to new state s) , and receive reward r
8: Q(s, a) ← (1 − α) Q(s, a) + α(r + γ maxa) Q(s) , a) ))
9: s ← s)
10: until the agent’s life is over
11: until Q(S, A) isn’t changing much any more, or we have run out of time
12: return Q(S, A) " As our approximation of Q∗ (S, A)
How does the agent decide what action to make? The algorithm will converge, slowly, to the
optimum if the action is picked entirely at random. Alternatively, you could pick the best action
possible for the state s, that is, use π ∗ (s), otherwise known as argmaxa Q∗ (s, a). Oh that’s right, we
don’t have Q∗ . Well, we could fake it by picking the best action we’ve discovered so far with our
(crummy) Q-table, that is, argmaxa Q(s, a).
That seems like a nice answer. But it’s got a problem. Let’s go back to our cockroach example.
The cockroach is wandering about and discovers a small candy. Yum! As the cockroach wanders
about in the local area, nothing’s as good as that candy; and eventually for every state in the local
area the cockroach’s Q table tells it to go back to the candy. That’d be great if the candy was
the only game in town: but if the cockroach just wandered a bit further, it’d discover a giant pile
of sugar! Unfortunately it’ll never find that, as it’s now happy with its candy. Recognize this
problem? It’s Exploration versus Exploitation all over again. If we use the best action a that we’ve
discovered so far, Q-learning is 100% exploitative. The problem is that the model-free version of the
algorithm, unlike the dynamic programming version, has local optima. We’re getting trapped in a
local optimum. And the solution is straight out of stochastic optimization: force more exploration.
We can do this by adding some randomness to our choices of action. Sometimes we do the best
action we know about so far. Sometimes we just go crazy. This approach is called "-greedy action
selection, and is guaranteed to escape local optima, though if the randomness is low, we may be
waiting a long time. Or we might do a Simulated Annealing kind of approach and initially just do
crazy things all the time, then little by little only do the best thing we know about.
Last, it’s fine to have α be a constant throughout the run. Though you may get better results if
you reduce α for those Q(s, a) entries which have been updated many times.
173
Generalization Believe it or not, there was a reason we covered all this. Reinforcement Learning
would be the end of the story except for a problem with the technique: it doesn’t generalize.
Ordinarily a learner should be able to make general statements about the entire environment based
on just a few samples of the environment. That’s the whole point of a learning algorithm. If you
have to examine every point in the space, what’s the point of using a learning algorithm? You’ve
already got knowledge of the entire universe.
Reinforcement Learning learns a separate action for every point in the entire space (every
single state). Actually it’s worse than that: Q-learning develops a notion of utility for every possible
combination of state and action. Keep in mind that in the Soccer Robot example, there were 100 states
and 6 actions. That’s a database of 600 elements! And that’s a small environment. Reinforcement
Learning doesn’t scale very well.
Many approaches to getting around this problem are basically versions of discretizing the space
to reduce its size and complexity. Alternatively you could embed a second learning algorithm —
typically a neural network — into the reinforcement learning framework to try to learn a simple set
of state action rules which describe the entire environment.
Another approach is to use a metaheuristic to learn a simple set of rules to describe the
environment in a general fashion. Such systems typically use an evolutionary algorithm to cut up
the space of states into regions all of which are known to require the same action. Then each rule is
simply of the form region description→action. Instead of having one rule per state, we have one rule
per region, and we can have as few regions as it takes to describe the entire space properly. We’ll
cover those next. But first...
A Final Derivation You can skip this if you like. The goal is to show where the magic equation
Q∗ (s, a) = R(s, a) + γ ∑ P(s) |s, a) max Q∗ (s) , a) )

s)
) a
came from. We’re going to go through the derivation of Q∗ in a very pedantic fashion. First, we
define Q∗ as telling us, for any given state s and action a, how good it would be to start in state s,
then perform action a, and then perform the smartest possible actions thereafter (that is, thereafter,
we use π ∗ (s) for all of our a). We can define Q∗ as the expected value, over all possible future strings
of states and actions, of the sum total reward we’d get for starting in s and doing a, then being
smart. Here’s how to write that:
∞
Q∗ (s, a) = E[ ∑ R(st , at )|s0 = s, a0 = a, at≥1 = π ∗ (st )]
t =0
There’s a problem. Imagine that there are two actions A and B, and if you always do action A,
regardless of your state, you get a reward of 1. But if you always do action B, you always get a
reward of 2. If our agent’s lifetime is infinite, both of these sum to infinity. But clearly B is preferred.
We can solve this by cutting down future rewards so they don’t count as much. We do this by
adding a multiplier 0 < γ < 1, raised to the power of t so it makes future rewards worth less. This
causes the sums to always be finite, and B’s sum to be higher than A’s sum.
∞
Q∗ (s, a) = E[ ∑ γt R(st , at )|s0 = s, a0 = a, at≥1 = π ∗ (st )] (3)
t =0
174
Now let’s pull our first actions s and a out of the sum. In the sum they’re known as s0 and a0 .
They’ll come out with their associated γ, which happens to be γ0 .
∞
Q∗ (s, a) = E[γ0 R(s0 , a0 ) + ∑ γt R(st , at )|s0 = s, a0 = a, at≥1 = π ∗ (st )]
t =1
From now on out, the goal is going to be to massage the stuff inside the expectation so that it
looks like the expectation in Equation 3 again. Let’s get going on that. Obviously γ0 = 1 so we can get
rid of it. Now there’s nothing in the expectation that R(s0 , a0 ) relies on so it can be pulled straight
out, at which time we can rename s0 and a0 back to s and a.
∞
Q∗ (s, a) = R(s, a) + E[ ∑ γt R(st , at )|s0 = s, a0 = a, at≥1 = π ∗ (st )]
t =1
Next comes the most complex part of the derivation. We’d like to get rid of the s0 and a0
still inside the expectation. So we’ll create a new state s) to be the next state s1 . But recall from
Figure 63, there are actually many possible states s)(1) , s)(2) , ... each with an associated probability
P(s)(1) |s, a), P(s)(2) |s, a), ... that the given s) state will be the one we wind up landing in after doing
action a in state s. So if we pull s0 out of the expectation, nothing in the expectation will reflect
this fact, and we’ll have to explicitly state that the old expectation has been broken into multiple
expectations, one per s) , and we’re adding them up, multiplied by the probabilities that they’d
occur. Here we go:
∞
Q∗ (s, a) = R(s, a) + ∑ P(s) |s, a) E[ ∑ γt R(st , at )|s1 = s) , at≥1 = π ∗ (st )]
s) t =1
Now we can change the inner sum back to t = 0, because there’s nothing inside the expectation
that relies on timestep 0 anymore. So inside the expectation we’ll just redefine t = 1 to be t = 0.
This will cause everything to be multiplied by one fewer γ so we’ll need to add a γ as well:
∞
Q∗ (s, a) = R(s, a) + ∑ P(s) |s, a) E[γ ∑ γt R(st , at )|s0 = s) , at≥0 = π ∗ (st )]
s) t =0
That γ isn’t dependent on anything, so we can pull it clear out of the expectation and the sum:
∞
Q∗ (s, a) = R(s, a) + γ ∑ P(s) |s, a) E[ ∑ γt R(st , at )|s0 = s) , at≥0 = π ∗ (st )]
s) t =0
Notice that inside the expectation we now have a new s0 but no a0 . We remedy that by breaking
our at≥0 up again. Instead of defining a0 to be π ∗ (s0 ), we’re going to invent a new symbol a) to
represent the action we perform when we’re in s) , that is, a) = a0 . This allows us to move the a)
definition outside of the expectation. But once again to do this we have to keep around the notion
that a) is the smartest possible action to perform when in a given s) . We do this by introducing the
operator max to select the a) that yields the highest possible expectation (that is, it’s the smartest
pick, and so is clearly π ∗ (s0 )):
∞
Q∗ (s, a) = R(s, a) + γ ∑ P(s) |s, a) max E[ ∑ γt R(st , at )|s0 = s) , a0 = a) , at≥1 = π ∗ (st )]
s)
) a t =0
175
And now the payoff for all this manipulation. Notice that the expectation (everything after the
max) now looks very similar to Equation 3. The only difference is that we’re using s) instead of s and
a) instead of a. This allows us to just say:
Q∗ (s, a) = R(s, a) + γ ∑ P(s) |s, a) max Q∗ (s) , a) )

s)
) a
Ta-da! A recursive definition pops out by magic!
10.2 Sparse Stochastic Policy Optimization

As mentioned before, the primary issue with reinforcement learning is that it constructs a unique
rule for every possible state. Q-learning is even worse, as it builds a table for every state/action
combination. If there are lots of states (and lots of actions), then there are going to be a lot of slots
in that table. There are a variety of ways to counter this, including simplifying the state space or
trying to use a learning method like a neural network to learn which states all have the same action.
Popular current methods include Ronald Williams’s R EINFORCE151 algorithms and Andrew Ng
and Michael Jordan’s P EGASUS,152 techniques collectively known as policy search.
We could also use metaheuristics to learn a sparse representation
of this rule space. The idea is to learn a set of rules, each of which
attach an action not to a single state but to a collection of states with e
c
some feature in common. Rather than have one rule per state, we
# "
search for a small set of rules which collectively explain the space in
some general fashion.
a
!
Imagine that states describe a point in N-dimensional space. For
example, in our soccer robot example, we might have four dimen-
d
! !"b
sions: ball size, ball position (including “not there”), goal size, and "
goal position (including “not there”). In the cockroach example, we
might have two dimensions: the x and y values of the grid location
of the cockroach. In the Tic-Tac-Toe example we might have nine Figure 65 A sparse version of
dimensions: each of the board positions. Given an N-dimensional the optimal policy for the cock-
space, one kind of rule might describe a box or rectangular region roach robot world, with five rules
(a...e). Compare to Figure 62.
in that space rather than a precise location. For example, here’s a The state marked • is covered by
possible rule for the cockroach robot: three different rules (a, c, and d),
with d being the most specific.
x ≥ 4 and x ≤ 5 and y ≥ 1 and y ≤ 9 → go up
Such a rule is called a classification rule, as it has classified (or labelled) the rectangular region
from $4, 1% and $5, 9% with the action “go up”. The rule is said to cover this rectangular region. The
objective is to find a set of rules which cover the entire state space and properly classify the states
in their covered regions with the actions from the optimal policy. For example, in Figure 65 we
have a small set of rules which collectively define exactly the same policy as shown in Figure 62.
If rules overlap (if the problem is over-specified), we may need an arbitration scheme. Were I
to hand-code such a ruleset, the arbitration scheme I’d pick would be based on specificity: rules
151 Ronald J. Williams, 1992, Simple statistical gradient-following algorithms for connectionist reinforcement learning,
in Machine Learning, pages 229–256.

152 This is a nontrivial paper to read. Andrew Ng and Michael Jordan, 2000, P EGASUS : A policy search method for large
MDPs and POMDPs, in Proceedings of the Sixteenth Conference on Uncertainty in Artificial Intelligence, pages 406–415.
176
covering smaller regions defeat rules covering larger regions. Figure 65 does exactly that. But the
methods discussed later use different approaches to arbitration.
There are two basic ways we could use a metaheuristic to learn rulesets of these kinds:
• A candidate solution (or individual) is a complete set of rules. Evolving rulesets is known as
Pitt Approach Rule Systems.
• An individual is a single rule: and the whole population is the complete set of rules. Evolving
individual rules and having them participate collectively is known as the Michigan Approach
to Learning Classifier Systems, or just simply Learning Classifier Systems (LCS).153
10.2.1 Rule Representation

State-action rules in Q-learning took the following form:
If I am in the following state... → Then do this...
The first part is the rule body, which defines the kinds of world states which would trigger the
rule. The second part is the rule head, which defines the action to take when the rule is triggered.
We can generalize the rule body in two different ways to cover more than one state. First, rule
bodies might require exact matches:
If I am in a state which exactly fits the following features...→ Then do this...
Or we can have rules which describe imprecise matches:
If I am in a state which sort of looks like this, even with a few errors...→ Then do this...
In the first case, we have the issue of under-specification: we need to make sure that for every
possible state, there’s some rule which covers that state. To guarantee this we might need to rely
on some kind of default rule which is assumed to match when no others do. Alternatively, the
algorithm might generate a rule on-the-fly, and insert it into the ruleset, to match a state if it
suddenly shows up.
In the second case, we don’t need to worry about under-specification, since every rule matches
every state to some degree. But we will need to define a notion of how well a rule matches. This is
known as a rule’s match score. The rule which the best match score might be selected.
In either case, we’ll still need to worry about over-specification, requiring an arbitration scheme.
Instead of specificity, the later methods use some combination of:
• The utility of the rule — essentially its Q-value, determined by the agent as it has tried out
the rule in various situations. Recall that utility is a measure of how often the rule led to high
rewards. Higher utility rules might be preferred over lower-utility rules.
• The variance in the rule’s utility: if the rule is consistent in yielding high rewards, it might be
preferred over more tenuous rules which occasionally get lucky.
153 Don’t confuse these with classification algorithms from machine learning, such as those mentioned in Section 9.1.
Those algorithms find classifications for whole regions of space based on provided samples in the space which have
been pre-labelled for them (part of an area called supervised learning). Whereas the metaheursistics described here find
classifications for regions based solely on reinforcement information gleaned while wandering about in the space.
177
• The error in the rule’s utility: the difference between the rule’s utility and the utilities of rules
which it leads to.
• The match score of the rule: rules more apropos to the current situation would be preferred
over ones whose bodies don’t match the situation very well.
Much of rule representation concerns itself with the rule body, which can take on many forms,
so it’s worth considering them:
Real-Valued or Integer Metric Spaces This state space is particularly common in Pitt-approach
rule systems, though it’s being increasingly studied in Michigan-approach methods too. There are
lots of ways you could describe the space, though boxes are the most common. Here are a few:
• Boxes We’ve seen these already.

Example: x ≥ 20 and x ≤ 30 and y ≥ 1 and y ≤ 9 → go up
Match Score: If the point’s in the box, it’s got a perfect match score (1.0 maybe?);
else perhaps its match score is equal to the percentage of dimensions
in whose ranges it lies. For example, $40, 5% is covered by y but not by
x, which might result in a match score of 0.5. Another approach: the
match score decreases with distance from the box boundary.
• Toroidal Boxes If your state space is bounded, a box could go off of one side and wrap
around to the other. Imagine if the space was toroidal in the x direction and bounded from 0
to 360. The rule below would be true either when 60 ≤ x ≤ 360 or 0 ≤ x ≤ 20 (assuming y is
in the right region). This isn’t totally nuts: it’s useful if x described an angle, for example.
Example: x ≥ 60 and x ≤ 20 and y ≥ 1 and y ≤ 9 → go up
Match Score: Same as regular boxes.
• Hyperspheres or Hyperellipsoids A rule might be defined as a point (the center of the
sphere) and a radius. Or a rule might be defined as a point and associated data describing a
rotated multidimensional ellipsoid (perhaps a covariance matrix like those used to describe
multidimensional Gaussian curves). Here’s an example of a simple hypersphere:
Example: If the state $ x, y, z% lies within a sphere centered at $4, 7, 2% and of
radius 9.3 → go up
Match Score: Same notion as regular boxes.
• Exemplars Specific points in the space which serve as examples for the regions around
them. A ruleset of exemplars divides the environment up into a Voronoi tessellation: regions
of space delimited by which exemplar each region is closest to.154 Such rules rely entirely on
match scores, so certain techniques (often Michigan approach methods) might not be able to
use them. You may think of exemplars as infinitely small hyperspheres.
Example: If the state is nearest to $4, 7, 2% → go up
Match Score: The further from the exemplar, the lower the match score.
154 After
Georgy Feodosevich Voronoi, 1868–1908, a Russian mathematician. Voronoi tesselations (sometimes called
“Voronoi diagrams”) are widely used in lots of areas of computational geometry, everything from graphics to wireless
networks to robotics. The notion of dividing space up by exemplars also forms the basis of the k-Nearest-Neighbor
(KNN) machine learning algorithm.
178
• Hyperplanes155 The rule cuts a plane through the space, dividing an area on which we
have an opinion from an area in which the rule has no opinion. Hyperplanes may likewise be
problematic for some Michigan approach methods.
Example: If 2.3x + 9.2y − 7.3z > 4.2 → go up
Match Score: If the point is on the matching side of the hyperplane, it matches
perfectly (or its match score improves if further away from the plane).
If the point is on the non-matching side of the hyperplane, its match
score is worse, but improves as it approaches the hyperplane.
Non-Metric Integer Spaces As we’ve seen earlier in the Section 4 (Representation), integer spaces
might describe metric spaces or simply define unordered sets of objects (0 = “red”, 1 = “blue”,
etc.). Integer-space rule bodies are no different. An unordered integer rule might look like this:
x = “red” and y = “soft” and z = “hollow” → go up
Here the rule, like exemplars, describes an exact point in the (unordered) space. A match score
might be defined in terms of the number of variables which exactly match the given state.
Unordered set rules might also have disjunctions:
x = “red” and y = “soft” and z = (“hollow” or “solid”) → go up
A disjunction would be considered a single condition, and it’d be true if any of its parts were true.
Boolean Spaces Though they have lately been generalized to other kinds of rules, Michigan
Approach classifier systems have traditionally focused on a single kind of rule: one involving
boolean conditions.
Because they’re so simple, boolean rules tend to take on a certain standard pattern: combinations
of “yes”, “no”, and “doesn’t matter”. Let’s say each state in your state space is described by three
boolean values, x, y, and z. Thus your space has eight states. A boolean rule over three dimensions,
might look like this:
x = 1 and y = 0 (and z doesn’t matter) → go up
In the parlance of Michigan Approach classifier systems, such a rule is usually written like this:
10# → go up
Note that the # sign means “this one doesn’t matter”. The more “doesn’t matter” dimensions in
the rule, the less specific. Match scores might again be defined in terms of the number of values
(that “matter”) which exactly match the state.
Could rule bodies be trees or graphs? More complex functions? Who knows?
155 There’sa clever way of converting hyperplanes into more complex subregions of space, called kernelization, a
technique made popular by Support Vector Machines (SVMs) in machine learning. I’ve not had much luck with
kernelization in the context of rule systems though.
179
10.3 Pitt Approach Rule Systems
The Pitt Approach156 applies an evolutionary algorithm to find a set of rules which best describes
the optimal policy. A candidate solution is simply a set of such rules. Section 4.5 introduced the
notions of rulesets popularly used in Pitt Approach rule systems and suggested approaches to
initializing, recombining, and mutating them. Here we will discuss a particularly well-known Pitt
Approach algorithm, SAMUEL.157
S AMUEL was developed by John Grefenstette, Connie Ramsey, and Alan Schultz at the Naval
Research Laboratory.158 The idea is to employ a Pitt Approach to optimizing rulesets as entire
candidate solutions in stochastic optimization, and to also use reinforcement learning ideas to
improve the rules within a candidate solution. S AMUEL traditionally uses a genetic algorithm,
but most any optimization method is plausible. All the actual magic is in the fitness assessment
function — where rule utilities are computed in addition to the fitness of the whole ruleset — and
in the breeding operators. S AMUEL iterates through four basic steps:
1. Each individual is tested n times and the results are used to update the utilities of its rules.
2. Using the updated utility information, each individual’s rules are improved in a special rule
mutation procedure.
3. Each individual is tested again some m additional times and the results are used to update
the fitness of the individual (ruleset) as a whole.
4. After all individuals have undergone the first three steps, we perform traditional evolutionary
algorithm style breeding and selection on the individuals based on fitness.
Fitness and Utility Assessment The two assessment steps (1 and 3 above) are nearly identical
except for the statistics they update: so we’ll treat them together here, and in fact Algorithm 124 is
used to describe both steps.
Both assessment procedures involve placing the agent in the world and having it follow the
policy as dictated by the ruleset being tested. As the agent is wandering about, we’ll need to decide
which action the agent will choose at any given step. This is first done by computing a match set
consisting of rules which best match the current state, that is, those with the highest match score.
Next, only the highest-scoring rules for each action are retained. S AMUEL then chooses a rule to
perform from the match set using some kind of score-based selection procedure. For example, we
might simply choose the rule with the highest score; or select with a probability proportional to
the rule’s score (as in fitness-proportionate selection, Algorithm 30). This two-level mechanism
156 Ken De Jong and students developed the Pitt Approach at the University of Pittsburgh. Hence the name.
157 S AMUEL is an acronym for Strategy Acquisition Method Using Empirical Learning. Yes, it’s pushing it. In reality,
Grefenstette, Ramsey, and Shultz were looking for a way to name the algorithm after Arthur Samuel, a famous machine
learning pioneer who (coincidentally I believe) died the same year as the seminal S AMUEL paper. While at IBM in
the 1950s, Arthur Samuel developed a program which learned on its own how to play checkers, and this program is
considered a major landmark in artificial intelligence history. Hmm, I seem to have a lot of footnotes about checkers....
S AMUEL was first defined in John Grefenstette, Connie Ramsey, and Alan Schultz, 1990, Learning sequential decision
rules using simulation models and competition, Machine Learning, 5(4), 355–381. Though you can get a roughly current ver-
sion of the manual online via CiteSeerx , presently at http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.48.9876
158 Where radar was invented!
180
(truncation followed by score-based selection) is intended to prevent large numbers of identical
crummy rules from being selected over a few high-quality ones.
The first fitness assessment procedure updates utility information about the rules. Recall that
Q-learning assumes that rewards occur throughout the agent’s life. In contrast, S AMUEL assumes
that rewards tend to happen at the end of an agent’s life. This leads to different strategies for
distributing rewards. In Q-learning, when a reward is received, it is stored in the Q-value for that
state-action combination; and later on when another state-action combination leads to this state,
the Q-value is then partially distributed to the earlier combination. We’ll see this assumption again
in Michigan Approach methods, in Section 10.4. But S AMUEL instead directly and immediately
distributes rewards to all state-action rules which led to the reward. Such rules are called active.
More specifically: if a rule contained an action which was used at some time in the past, prior to a
reward r appearing, then when r is finally received, the utility of the rule is updated as:
Utility( Ri ) ← (1 − α) Utility( Ri ) + αr
S AMUEL also maintains an approximation of the variance of the utilities of each rule because
we want to have rules which both lead to high rewards and are consistent in leading to them. Each
time the utility is updated, variance in utility is also updated as:
UtilityVariance( Ri ) ← (1 − α) UtilityVariance( Ri ) + α(Utility( Ri ) − r )2
Finally, S AMUEL uses this information to build up a “quality” of sorts of each rule, called the
rule’s strength,159 which is a combination its utility and utility variance. Strength affects how likely
the rule is to be mutated later on.
Strength( Ri ) ← Utility( Ri ) + γ UtilityVariance( Ri )
We commonly set γ to a low value less than 1, as utility is more important than variance.
Distributing reward evenly among all rules is an odd choice. I would have personally dis-
tributed so that later rules received more reward than earlier rules. Interestingly, S AMUEL maintains
information about how long ago a rule was active, though it uses it only to determine which rules
to delete. This value is called the activity level of a rule. Rules start with an activity level of 12 , and
are updated each time the agent performs an action. Rules which had that particular action in their
heads are increased like this:
Activity( Ri ) ← (1 − β) Activity( Ri ) + β
Given an 0 ≤ β ≤ 1, this has the effect of shifting a rule’s activity towards 1 when the rule’s
action is chosen. Rules without that action in their heads have their activity levels decreased:
Activity( Ri ) ← δ Activity( Ri )
for 0 ≤ δ ≤ 1. This has the effect of slowly decreasing the rule’s activity level towards zero.
The second assessment procedure in S AMUEL is used to compute the fitness of the entire
individual (the ruleset). This is simply defined as the sum of rewards received by the individual
during testing. The following algorithm describes both fitness procedures: the particular procedure
being done (utility or fitness) is determined by the dofitness variable.
159 Not to be confused with Pareto “strength” (Section 7.3).
181
Algorithm 124 S AMUEL Fitness Assessment
1: S ← individual being assessed
2: α ← learning and decay rate
3: β ← activity level increase rate
4: γ ← how much variance to include
5: δ ← activity level decay rate
6: dofitness ← are we assessing to compute fitness (as opposed to rule strength)?
7: n ← number of times to test the agent
8: f ←0
9: R ← { R1 , ..., Rl } rules in the ruleset of the individual S
10: for n times do
11: s ← an initial state of agent
12: Z ← {} " Active Rule Set
13: for each rule Ri ∈ R do " All rules which were in an action set this time around
14: Activity(Ri ) ← 0.5
15: repeat
16: for each rule Ri ∈ R do " No matter how badly they match the state
17: ComputeMatchScore(Ri , s)
18: N ← all actions which appear in the head of any rule in R
19: M ← {} " Match Set
20: for each action Nj ∈ N do " Find the highest-scoring rule for each action
21: R) ⊆ R ← all rules in R whose heads are action Nj
22: M ← M ∪ { the rule Ri) ∈ R) whose match score is highest }
23: R a ← SelectWithReplacement(M) " Select among the highest-scoring rules
24: A ⊆ R ← all rules whose heads (actions) are the same as the head of R a " Action Set
25: for each rule Ai ∈ A do " Increase activity
26: Activity(Ai ) ← (1 − β) Activity(Ai ) + β
27: if Ai ∈/ Z then
28: Z ← Z ∪ { Ai }
29: for each rule Ri ∈ R − A do " Decrease activity
30: Activity(Ri ) ← δ Activity(Ri )
31: Perform action R a , transitioning to a new state s " Notice no reward
33: r ← cumulative reward (assessment) of the agent " Ah, here’s the reward. Only at the end.
34: if dofitness is false then " We’re doing runs to update the strengths of the rules
35: for each rule Zi ∈ Z do
36: Utility(Zi ) ← (1 − α) Utility(Zi ) +αr
37: UtilityVariance(Zi ) ← (1 − α) UtilityVariance(Zi ) +α(Utility(Zi ) − r )2
38: Strength(Zi ) ← Utility(Zi ) − γ UtilityVariance(Zi )
39: else " We’re doing runs to update fitness
40: f ← f +r
41: if dofitness is true then
42: fitness of S ← f
182
Mutation S AMUEL has two mutation steps, each following one of the assessment steps. After
the first assessment procedure (which determines rule strength), the rules in the individual are
modified. Hopefully this improves the individual for the second fitness assessment (whose purpose
is to compute the actual fitness of the individual). After the second fitness procedure, we do regular
breeding of the population with more bulk-style, traditional operations.
Let’s start with the first mutation step: improving the rules. S AMUEL performs any of the
following mutations on the individual to try to improve it for the second stage:
• Rule Deletion If a rule is sufficiently old (brand new rules are never deleted), has a
sufficiently low activity value (it’s not fired recently), or its strength is sufficiently low, or if
the rule is subsumed by another rule with greater strength, then the rule is a candidate for
deletion. We may also delete a few rules randomly. It’s up to you to decide these thresholds
and how many deletions occur. We say that a rule A is subsumed by another rule B if every
state that A covers is also covered by B, and B covers some additional states as well, and the
two rules have the same actions in their heads.
• Rule Specialization If a rule is not very strong and covers a large number of states, it is a
candidate for specialization because it may be crummy because of the large region it’s covering.
We add to the ruleset a new rule subsumed by the old rule (and thus more specific) and which
has the same action in its head. The original rule is retained. For example, the rule
Might be specialized to
x = 5 and y ≥ 6 and y ≤ 9 → go up
• Rule Generalization This is the opposite of rule specialization. If a rule is very strong and
covers a small number of states, it is a candidate for generalization because it might do well
with more states. We add to the ruleset a new rule which subsumes the old rule (and thus is
more general) and has the same action in its head. The original rule is retained.
• Rule Covering Covering is similar to generalization, but is based on information we

gleaned from the assessment process. Let’s say that during assessment we discovered that a
certain rule had often fired but was fairly consistent in not completely matching the state. For
example, returning to our rule
Imagine that this rule had been selected a number of times when y = 4, x = 6. Obviously
x = 6 is out of bounds for the rule, but the y = 4 match was good enough, and the rule was
strong enough, for it to win even with only a partial match. Rule covering would select this
rule and create a new one more likely to match, for example:
The original rule is retained.
183
• Rule Merging If two rules are sufficiently strong, share the same action in their heads, and
overlap sufficiently in the number of states they cover, they’re candidates for merging into a
single rule which is the union of them. The original rules are retained.
Notice that all these mutation mechanisms are directed, that is, they’re explicitly exploitative,
aimed at pushing the rules so that they perform better next time. For this reason, John Grefenstette
refers to this mutation step as Lamarckian (see Section 4.3.4) — it improves the individuals during
the course of assessment.
The remaining mutation operators occur during breeding just like any other evolutionary
algorithm, and have more of an explorative nature to them:
• Plain Old Mutation Make random mutations to some rules. The original rules are not
retained. This is the more explorative mutation.
• Creep Mutation160 Make a very small, local random change to a few rules. The objective
here is to push a little bit for hill-climbing.
Recombination Section 4.5.5 mentioned various approaches to crossing over rulesets. S AMUEL
offers other possibilities:
• A version of Uniform Crossover Some n times, the two individuals trade a rule at random.
• Clustered Crossover From the fitness assessment procedure we gather some statistics:
specifically, we want to know which sequences of rules led to a reward. From this we identify
pairs of rules which often led to a reward when they both appeared in a sequence. We then do
a uniform crossover, but at the end try to ensure that these pairs don’t get split up: if one rule
winds up in individual A and the other in individual B, we move one of the two rules to the
other individual (swapping over some other rule instead). The idea is to recognize that there
is very strong linkage among rules in rulesets, and we want to cross over whole teams of rules
which have performed well as a group.
Notice that both of these recombination operators don’t change the size of either ruleset. Nor
do the mutation operators during breeding. S AMUEL appears to restrict ruleset size changes to the
exploitative “Lamarckian” mutation operators which occur after the first assessment procedure.
Selection You can use any old fitness-based selection procedure. Though S AMUEL traditionally
uses an odd combination of truncation selection and Stochastic Universal Sampling. Specifically,
we compute the mean fitness over the whole population, as well as the variance in the fitness. We
then update a baseline fitness as follows:
baseline ← (1 − υ) baseline + υ(mean fitness − ψ variance in fitness)
... where 0 ≤ υ ≤ 1, and ψ is a parameter indicating how important variance is. Once we have our
baseline fitness, the only individuals which are even considered for selection are those whose fitness
is higher than the baseline. We then use a standard selection procedure (S AMUEL used Stochastic
Universal Sampling) to select among those individuals.
In truth, I wonder if just doing plain-old truncation selection would do just as well.
160 My vote for creepiest mutation name.
184
Initialization There are lots of ways to initialize the ruleset. In S AMUEL three are common:
• Create a set of random rules.
• Seed the rules in each individual with rules you believe to be helpful to the agent.
• Perform adaptive initialization. Each individual starts with a set of rules that are totally
general, one for each possible action:
In all cases → go up In all cases → go down ...etc...
Run this for a while to get an idea of the strength of each rule. As you’re doing this, apply
a fair number of Rule Specialization operators, as described earlier, to make these general
rules more specific. The idea is to gracefully let S AMUEL find good initial operators based on
a bit of initial experience in a sandbox.
Self-Adaptive Operators S AMUEL has an optional gimmick for adjusting the probability that
various mutation operators will occur (particularly the “Lamarckian” ones). Each individual
contains its own operator probabilities. Let’s say that P(Oi , Ij ) is the probability that operator Oi is
performed on individual Ij . This probability is stored in individual Ij itself, and children receive
the same set of probabilities that their parents had. Each timestep all the operator probabilities in
all individuals are decreased like this:
P(Oi , Ij ) ← (1 − τ ) P(Oi , Ij )
... where 0 ≤ τ ≤ 1. This eventually pushes the probabilities towards 0. But when an individual
is mutated or crossed over using an operator, the probability of that operator is increased for the
resulting individual(s), perhaps something like:
P(Oi , Ij ) ← (1 − τ ) P(Oi , Ij ) + τ
This pushes this probability, eventually, towards 1.

This is an example of self-adaptive operators, where the individuals contain their own mutation
and crossover probabilities. Self-adaptive operators have been around for a long time, since early
work in Evolution Strategies. But in my personal experience they’re finicky.161 I wouldn’t bother.
10.4 Michigan Approach Learning Classifier Systems

After John Holland162 developed the Genetic Algorithm around 1973, he turned his attention to
a related topic: how to use an evolutionary process to discover a set of rules which describe, for
each situation an agent finds himself in, what to do in that situation. I think Holland pitched
it more generally than this — as a general machine learning classifier rather than one used for
161 My proposed dissertation work was originally going to be using self-adaptive operators. Let’s just say I wound up
doing something else.
162 John Holland is at the University of Michigan. Hence the name. Holland’s earliest work on the topic is John Holland,
1975, Adaptation in Natural and Artificial Systems, University of Michigan Press. But the notion of learning classifier
systems weren’t formalized until a later paper, John Holland, 1980, Adaptive algorithms for discovering and using
general patterns in growing knowledge bases, International Journal of Policy Analysis and Information Systems, 4(3), 245–268.
185
agent actions — and this is where the name Learning Classifier Systems (LCS) came from. Rather
than having individuals being whole solutions (rulesets), Holland envisioned a population of
individual rules which would fight for survival based on how effective they were in helping the
classifier as a whole. Thus, like Ant Colony Optimization, Learning Classifier Systems have a very
one-population coevolutionary feel to them.
Holland’s original formulation was somewhat baroque. Since then, Stewart Wilson has created
a streamlined version called the Zeroth Level Classifier System (ZCS).163 ZCS is a steady-state
evolutionary computation technique. The evolutionary computation loop iterates only occasionally.
Instead, most of the time is spent updating the fitness values of the entire generation based on their
collective participation, as rules, in a reinforcement learning setting. Then after a while a few new
rules are bred from the population and reinserted into it, displacing some existing low-fitness rules.
ZCS maintains a population of sparse if→then rules. Each rule is associated with a current
fitness which reflects the utility of the rule. To test the rules, the agent is placed in a starting state,
and then begins performing actions chosen from the population. This is done by first selecting
all the rules which cover the current state of the agent. This set of rules forms the match set M. If
there is more than one such rule, ZCS’s arbitration scheme selects from among the match set using
a fitness-based selection method (traditionally fitness-proportionate selection).
One way in which ZCS differs from S AMUEL is that it expects a complete match rather than
allowing partial matches. Match scores are never used. If the match set is in fact empty — not
a single rule covers the current state — ZCS creates a random rule which covers the state (and
possibly others), and which has a random action. The fitness of the rule is set to the average fitness
of the population at present. ZCS then marks an existing rule for death in the population and
replaces it with this new rule. Rules are usually marked for death via a fitness-based selection
method, tending to select less-fit rules more often.
Once ZCS has a winning rule, it creates a subset of the match set called the action set A,
consisting of all the rules whose head was that action. The action is performed, and the agent
receives a reward r and transitions to a new state s) , at which point ZCS constructs the next match
set M) and action set A) . Each rule Ai ∈ A then has its fitness updated as:
1
Fitness( Ai ) ← (1 − α) Fitness( Ai ) + α (r + γ ∑ Fitness( A)j )) (4)
|| A|| A) ∈ A)
j
Look familiar? Hint: let’s define a function G, consisting of the combined fitness (utility) of
all the rules in the present action set A. That is, G ( A) = ∑i Fitness( Ai ). Equation 4 above would
result in the equivalent equation for G:
G ( A) ← (1 − α) G ( A) + α(r + γ G ( A) ))
Compare this to Equation 2. Unlike S AMUEL, ZCS updates utility (ZCS’s rule fitness) in basically
a Q-learning fashion. ZCS also punishes rules for not getting picked (that is, the rules in M − A).
Let B = M − A. Then the fitness of each rule Bi ∈ B is decreased as:
Fitness( Bi ) ← β Fitness( Bi )
This has basically the same effect as evaporation did in Ant Colony Optimization. β can be a
value between 0 and 1, and shouldn’t be very large. All told, the algorithm for updating fitnesses
in the match set is:
163 Introduced in Stewart Wilson, 1994, ZCS: A zeroth level classifier system, Evolutionary Computation, 2(1), 1–18.
186
Algorithm 125 Zeroth Classifier System Fitness Updating
1: M ← previous match set
2: M ) ← next match set " Unused. We keep it here to be consistent with Algorithm 131.
3: A ← previous action set
4: A) ← next action set
5: r ← reward received by previous action
7: β ← evaporation constant " 0 < β < 1. Make it large.
9: for each Ai ∈ A do
10: Fitness( Ai ) ← (1 − α) Fitness( Ai ) + α || A1 || (r + γ ∑ A)j ∈ A) Fitness( A)j ))
11: B ← M−A
12: for each Bi ∈ B do
13: Fitness( Bi ) ← β Fitness( Bi )
Because ZCS uses fitness as utility, when ZCS produces children as a result of steady-state
breeding, it needs to assign them an initial fitness: otherwise they would never even be considered
for match sets. To this end, half the fitness of each parent is removed from the parent and added into
each child (because we want to approximately maintain the sum total fitness in our population):
Algorithm 126 Zeroth Classifier System Fitness Redistribution

1: Pa , Pb ← parents
2: Ca , Cb ← children
3: crossedover ← are the children the result of crossover?
4: if crossedover = true then

5: Fitness(Ca ), Fitness(Cb ) ← 14 ( Fitness(Pa ) + Fitness(Pb ) )
6: else
7: Fitness(Ca ) ← 12 Fitness(Pa )
8: Fitness(Cb ) ← 12 Fitness(Pb )
1
9: Fitness(Pa ) ← 2 Fitness(Pa )
1
10: Fitness(Pb ) ← 2 Fitness(Pb )
Now we can examine the top level ZCS loop. The loop has two parts:
1. We update the utilities (fitnesses) of the rules by testing them with the agent: we repeatedly
create a match set, pick an action from the match set, determine the action set, perform the
action and receive reward, and update the fitness values of the rules in the match set. Fitness
values are updated with Algorithm 125.
2. After doing this some n times, we then perform a bit of steady-state breeding, producing
a few new rules and inserting them into the population. The fitness of the new children is
initialized using Algorithm 126.
187
Algorithm 127 The Zeroth Level Classifier System (ZCS)
2: f ← fitness value to be assigned to initial population members " Can be whatever. Say, 1.
3: n ← agent runs per evolutionary loop " Make it large.
4: p ← probability of crossover occurring " Make it small.
5: P ← {} " Here P means a population (as usual), not the P(s) |s, a) distribution from Q-learning
7: P ← P ∪ {new random individual with initial fitness of f }
8: repeat " First we do the reinforcement stage to build up fitness values
9: for n times do
10: s ← an initial state of agent
11: r←0
12: M ← {}
13: A ← {}
14: repeat
15: M) ⊆ P ← match set for state s " That is, all Pi ∈ P which cover s
16: if M) is empty then " Oops, nothing’s covering s, make something
17: M) ← M) ∪ {new random individual covering s, fitness set to the population mean}
18: Individual Pc ← SelectForDeath(P)
19: P ← ( P − { Pc }) ∪ M)
20: a ← best action from M) " The action of the winner of SelectWithReplacement(M) )
21: A) ⊆ M) ← action set for action a " That is, all M)j ∈ M whose action is a
22: UpdateFitnesses with M, M) , A, A) and r
23: Have agent perform action a, resulting in new reward r and transitioning to new state s
24: M ← M)
25: A ← A)
27: UpdateFitnesses with M, M) , A, {} and r " Final iteration. Note M = M) , and A) = {}
28: Parent Pa ← SelectWithReplacement(P) " And now we begin the breeding stage
30: Child Ca ← Copy(Pa )
31: Child Cb ← Copy(Pb )
33: Ca , Cb ← Crossover(Ca , Cb )
34: RedistributeFitnesses(Pa , Pb , Ca , Cb , true)
35: else
36: RedistributeFitnesses(Pa , Pb , Ca , Cb , false)
40: Individual Pe ← SelectForDeath(P) " Pd must be 4= Pe
41: P ← ( P − { Pd , Pe }) ∪ {Ca , Cb }
42: until we have run out of time
43: return P
188
The parameter n specifies the number of fitness updates performed before another iteration of
steady-state evolution. If n is too small, the algorithm starts doing evolution on sketchy information,
and becomes unstable. If n is too large, the algorithm wastes time getting very high-quality fitness
information when it could be spending it searching further. Usually, n needs to be large.
In ZCS, crossover is optional. This is of course the case in many algorithms, but in ZCS it’s
particularly important because crossover is often highly destructive. The parameter p reflects how
often crossover is done in creating children (usually not often). If crossover occurs, the redistributor
is informed so as to average out the fitness values between them.
The ZCS algorithm is the first metaheuristic covered so far which doesn’t return a “best result”:
rather the entire population is the result. The population itself is the solution to the problem.
The XCS Algorithm Building on ZCS, Stewart Wilson developed a next-generation version which
he called XCS.164 XCS has since gone through a number of iterations, including additions from Pier
Luca Lanzi and Martin Butz. Basically XCS differs from ZCS in four primary places:
• How the action is selected
• How UpdateFitnesses is performed
• How SelectWithReplacement is done in the evolutionary portion of the algorithm
• How RedistributeFitnesses is performed
The big change in these algorithms is that XCS doesn’t use the Q-value as the fitness. Instead,
in addition to a fitness value, each individual also has a separate, explicit utility measure and a
utility error165 measure. The utility measure now holds the rough notion of Q-value. The utility
error measure is approximately the average difference between the current utility of the rule and
the current utility of the rules in the next time step. I say approximately because the utility error
measure (like the utility measure and also fitness) uses the 1 − α trick to “fold in” newer results;
thus recent utility errors count more than older ones. Ideally as rules get more effective, and the
utility gets more accurate, the utility error should go down. The fitness is based on the normalized
utility error with regard to errors of other members of the action set: lower error, better fitness.
Picking an Action XCS picks an action from the match set M by first determining the “best”
action in M. To do this it gathers all the rules in M which propose the same action. XCS then adds
up their utilities, probabilistically weighted by their fitnesses (fitter rules get to contribute more to
the utility of the action).
164 XCS doesn’t appear to stand for anything! The earliest version of the algorithm appeared in Stewart Wilson, 1995,
Classifier fitness based on accuracy, Evolutionary Computation, 3(2), 149–175.
XCS is complex. If there was an algorithm in which I made several mistakes (and I’m sure they’re there!) this would
be the one. For a more accurate description of the algorithm, see Martin Butz and Stewart Wilson, 2001, An algorithmic
description of XCS, in Advances in Learning Classifier Systems, volume 1996/2001, pages 267–274, Springer. Much of the
code in these lecture notes was derived from this paper. Note that my version has some simplifying syntactic changes
(no “prediction array” for example) but it should operate the same (knock on wood).
165 XCS calls these the prediction and prediction error of the rule.
189
Algorithm 128 XCS Fitness-Weighted Utility of an Action
1: M ← match set
2: Ni ← action
3: R ⊆ M ← all rules in M whose heads are Ni

4: if ∑r∈ R Fitness(r) 4= 0 then
∑ ( Utility(r) × Fitness(r) )
5: return r∈ R
∑r∈ R Fitness(r)
6: else
7: return 0
Now we can determine which of the actions is the “best one”:

Algorithm 129 XCS Best Action Determination
1: M ← match set
2: N ← all actions which appear in the head of any rule in M
3: Best ← !
4: bestc ← 0
5: for each action Ni ∈ N do
6: c ← XCS Fitness-Weighted Utility of action Ni
7: if Best = ! or c > bestc then
8: Best ← Ni
9: bestc ← c
10: return Best
Now we either pick a random action (with % probability), or we choose our “best” action. This
approach should look familiar: it’s once again %-greedy action selection, just like in Q-learning.
This was first proposed for XCS by Pier Luca Lanzi.166
Algorithm 130 XCS Action Selection

1: M ← match set
2: % ← exploration probability "0≤%≤1
3: N ← all actions which appear in the head of any rule in M

4: if % ≥ random number chosen uniformly from 0.0 to 1.0 inclusive then
5: return a member of N chosen uniformly at random
6: else
7: return the action provided by XCS Best Action Determination given M and N
166 Pier Luca Lanzi, 1999, An analysis of generalization in the XCS classifier system, Evolutionary Computation, 7(2),
125–149.
190
Updating Fitness During testing we no longer have just a fitness to update: we’ll need to update
all three elements: the utility, the utility error, and the fitness. The utility is updated Q-style:
Utility( Ai ) ← (1 − α) × Utility( Ai ) + α(r + γb)
What is b? It’s the XCS Fitness-Weighted Utility (Algorithm 128) of the best action (Algorithm
129) the next time around — so you’ll need to delay fitness updating of this iteration until you gone
one more iteration. Again, compare this to Equation 2.
The utility error is updated similarly, by rolling in the new error computed by subtracting the
utility from the likely best utility of the next action set:
UtilityError( Ai ) ← (1 − α) × UtilityError( Ai ) + α ||b − Utility( Ai )||
To compute the fitness, we first convert the error into an “accuracy” ai . If the error is greater
than some small value e, the accuracy ai is considered to be perfect, that is, 1. Otherwise, the
6 7β
e
accuracy ai is set to δ UtilityError (A )
. Finally the accuracy is rolled into the fitness:
i
ai
Fitness( Ai ) ← (1 − α) × Fitness( Ai ) + α
∑ Aj ∈ A a j
There’s no evaporation. Here’s the algorithm in full:
Algorithm 131 XCS Fitness Updating

1: M ← previous match set " Note: for the final iteration of the ZCS/XCS top loop, M = M)
)
2: M ← next match set
4: A) ← next action set " Unused. We keep it here to be consistent with Algorithm 125.
6: e ← the highest error in utility that should still warrant full fitness
8: β ← fitness adjustment parameter " β>1
10: δ ← fitness adjustment parameter " Presumably 0 ≤ δ ≤ 1. I’m guessing 1 is fine.
11: n ← the action returned by XCS Best Action Selection on M

12: b ← the XCS Fitness-Weighted Utility of action n
13: !a ← $ a1 , ..., a|| A|| % vector of accuracies, one per rule in A
14: for each rule Ai ∈ A do
15: Utility(Ai ) ← (1 − α) × Utility(Ai ) +α(r + γb)
16: UtilityError(Ai ) ← (1 − α) × UtilityError(Ai ) + α ||b − Utility(Ai )||
17: if UtilityError(Ai ) > e then " Convert error into “accuracy” (big error, low accuracy)
4 5β
e
18: ai ← δ
UtilityError(Ai )
19: else
20: ai ← 1 " Why it’s not ai ← δ I have no idea
ai
22: Fitness(Ai ) ← (1 − α) Fitness(Ai ) + α " Normalize the accuracies
∑ Aj ∈ A a j
191
Redistributing Fitness In addition to fitness, XCS now also needs to redistribute utility and
utility error. And unlike ZCS, rather than redistribute fitness from the parents, XCS just cuts down
the fitness of the child. Specifically:
Algorithm 132 XCS Fitness Redistribution

1: Pa , Pb ← parents
2: Ca , Cb ← children
3: ν ← fitness cut-down " Use 0.1
4: crossedover ← are the children the result of crossover?
5: if crossedover = true then

6: Fitness(Ca ), Fitness(Cb ) ← ν 14 (Fitness(Pa ) + Fitness(Pb ))
7: Utility(Ca ), Utility(Cb ) ← 14 (Utility(Pa ) + Utility(Pb ))
8: UtilityError(Ca ), UtilityError(Cb ) ← 14 (UtilityError(Pa ) + UtilityError(Pb ))
9: else
10: Fitness(Ca ) ← ν 12 Fitness(Pa )
11: Fitness(Cb ) ← ν 12 Fitness(Pb )
12: Utility(Ca ) ← 12 Utility(Pa )
13: Utility(Cb ) ← 12 Utility(Pb )
14: UtilityError(Ca ) ← 12 UtilityError(Pa )
15: UtilityError(Cb ) ← 12 UtilityError(Pb )
Performing SelectWithReplacement SelectWithReplacement is not performed over the whole

population as it was in ZCS. Rather, it’s just performed over the action set. That is, lines 28 and 29
of Algorithm 127 should look like this:
Parent Pa ← SelectWithReplacement(A)
Parent Pb ← SelectWithReplacement(A)
Other Gizmos To this basic algorithm, XCS normally adds some other gizmos. First, there’s the
notion of microclassifiers. XCS considers each individual not just as one rule, but actually as a
whole lot of rules that are exactly the same. This is done by including with each individual a count
variable which indicates how many “copies” of the rule are considered to be in the individual.
When we do fitness updating (Algorithm 131), the very last line includes this count variable so that
each of those “embedded” rules get a voice:
ai × Count(Ai )
Fitness(Ai ) ← (1 − α)Fitness(Ai ) + α
∑ A j ∈ A a j × Count(A j )
Counts also figure when we’re creating new rules or selecting rules for deletion. If we create a
new rule, we check first to see if it’s identical to an existing rule. If so, the existing rule has its count
increased, and the new rule isn’t actually added to the population. When we delete a rule, and its
count is higher than 1, we just decrease the count and retain the rule; only when its count is 1 do
we delete it. Note that this could result in the population size changing a bit. Notionally this gizmo
is a niching method which prevents large numbers of copies of the same rule from dominating the
population. But it might also have the effect of making long-established rules harder to get rid of.
192
Because initial fitness and utility is arbitrarily set, XCS also grants new rules a bit of leeway, to
give them a chance to get their utilities and utility errors ramped up. This is done by maintaining
an experience counter for each rule which is incremented each time that rule appears in an action
set. The learning rate is decreased little by little until the experience exceeds 1/α, at which point
the learning rate is α thereafter.
Putting this all together, we can extend the XCS Fitness Updating algorithm (Algorithm 131) to
include these additional gizmos:
Algorithm 133 XCS Fitness Updating (Extended)
1: M ← previous match set " Note: for the final iteration of the ZCS/XCS top loop, M = M)
)
2: M ← next match set
4: A) ← next action set " Unused. We keep it here to be consistent with Algorithm 125.
6: e ← the highest error in utility that should still warrant full fitness
8: β ← fitness adjustment parameter " β>1
10: δ ← fitness adjustment parameter " Presumably 0 ≤ δ ≤ 1. I’m guessing 1 is fine.
11: n ← the action returned by XCS Best Action Selection on M

12: b ← the XCS Fitness-Weighted Utility of action n
13: !a ← $ a1 , ..., a|| A|| % vector of accuracies, one per rule in A
15: Experience(Ai ) ← Experience(Ai ) +1
1
16: α) ← max( , α)
Experience(Ai )
17: Utility(Ai ) ← (1 − α) ) × Utility(Ai ) +α) (r + γb)
18: UtilityError(Ai ) ← (1 − α) ) × UtilityError(Ai ) +α) ||b − Utility(Ai )||
19: if UtilityError(Ai ) > e then " Convert error into “accuracy” (big error, low accuracy)
4 5β
e
20: ai ← δ )
UtilityError(Ai
21: else
22: ai ← 1 " Why it’s not ai ← δ I have no idea
ai × Count(Ai )
24: Fitness(Ai ) ← (1 − α) Fitness(Ai ) +α
∑ A j ∈ A a j × Count(A j )
The big changes are on lines 15, 16, and 24.

Finally, XCS has optional subsumption procedures: it checks for a subsumed rule whose covered
states are entirely covered by some other rule which is both reasonably fit and sufficiently old. The
goal is, once again, to force diversity and eliminate redundancy. Subsumption could show up in
two places. First, when a brand-new rule is created, XCS may refuse to include it in the population
if it’s subsumed by some other rule; instead, the subsuming rule has its count increased by one.
Second, after building an action set A, XCS could check A to see if any rules subsume any others. If
so, the subsumed rules are removed from the population.
193
10.5 Is this Genetic Programming?
In some important sense, policies are programs which control
Room B Room C
agents. These programs consist of if→then rules where the if side
Actions: Actions: Switch
consists of the current state of the world. Even without control Go to A Go to B (Initially Off)
structures, this is often a lot more sophisticated than the lion’s Go to C Flick Switch
share of “programs” that tree-structured or machine-code genetic Room A
programming develops (see Sections 4.3 and 4.4). But is this suf- Actions:
ficient to be called “programming”? Go to B
Exit Door
Well, in lots of environments, you need more than just the state
of the world to decide what to do. You also need a memory where Door
you store some form of information gleaned from the history of (Initially Closed)
what’s happened. That memory is typically called the internal Figure 66 A robot world with three
state of the agent (as opposed to the world state, or external state). rooms, a door, and a switch. avail-
Consider Figure 66 at right. The robot starts in room A and able actions for each room are shown.
wants to go out the door. We would like to develop a policy that The robot can only exit if the door is
opened. Flicking the switch opens
enables the robot to go to room C, flick the switch (which opens
the door.
the door), return to A, and go out the door. The policy might be:
In A and door closed → go to B

In B → go to C
In C and switch off → flick switch
In C and switch on → go to B
In B → um....
In A and door open → go out the door
The problem is that we already have a rule for B! Go to C. We need two rules for B: if I’m headed
to flick the switch, go to C, but if I’m headed out the door, go to A. Trouble is, in room B we have
nothing to go on, no external state information, which can help us distinguish these features. The
two B situations are aliased: they require different actions but exhibit the same external state.
We need some memory: specifically, we need memory of whether we flicked the switch or not.
Let’s give the agent a single bit of memory. Initially the bit is 0. Now we might construct this policy:
In A and door closed → go to B

In B and memory bit is 0 → go to C
In C and switch off → flick switch and set memory bit to 1
In C and switch on → go to B
In B and memory bit is 1 → go to A
In A and door open → go out the door
Problem solved! Here’s the thing: by adding a single bit of memory, we’ve potentially doubled
our state space. A single bit isn’t too bad, but several bits and we radically increase the complexity
of our world. Techniques for handling these issues are fairly cutting-edge. I personally view policy
optimization methods as the closest thing we have to successful “genetic programming” at present:
but we’re still a long ways from true automatic programming. Your job is safe.
194
11 Miscellany
Always the most interesting section of a book.167
11.1 Experimental Methodology

11.1.1 Random Number Generators, Replicability, and Duplicability
Random Number Generators Metaheuristics employ randomness to some degree. Like all
stochastic techniques, the validity of your results may rely on the quality of your random number
generator. Unfortunately, there are a lot of very very bad random number generators in common use.
Many of the more infamous generators come from a family of linear congruential random number
generators, where the next random number is a function of the previous one: xt+1 = ( axt + c)
mod m. The values for a, c, and m must be very carefully chosen in order for this generator to be
even adequate to use. But bad choices of these constants have led to some truly infamous results.
The R ANDU generator, for example, ruined experimental results as far back as the 1960s. A mistake
in the ANSI C specification led to the propagation of a horrible generator in C and C++’s rand()
function even to this day. And Java’s java.util.Random produces such non-random results that
there’s an entire web page devoted to making fun of it.168 When I examine new Java metaheuristics
toolkits, the first thing I check is whether they’re using java.util.Random or not.
The revelation of a poor generator has cast doubt on more than one research paper in the
literature. You ought to pick a high-grade generator. My own personal choice is the Mersenne
Twister, a highly respected generator with very good statistical properties and an ultra-long period
(the amount of time before it starts repeating its sequence), but there are other very good ones out
there as well.
Generators need to be seeded and used properly. Too many times have I seen beginners
repeatedly instantiating a new java.util.Random instance, generating one integer from it, then
throwing it away, seemingly blissfully unaware that this is grotesquely nonrandom. This awful
approach gives you a sequence of numbers loosely following your computer’s wall clock time. The
a good way to use a random number generator in your experiments is:
1. Choose a very high grade random number generator.
2. Pick a unique seed for each and every experimental run you do.
3. Seed your generator based on the seed you picked.
4. If you’re using a language like Java in which generators are objects, create only one generator
per experimental run and continue to use it throughout the run, never creating a new one.
Unless you know exactly what you’re doing, it’d be wise to not deviate from this procedure.
Replicability When you perform your experiments and write them up for a conference or journal,
you must strive for replicability. You should report your results in such a way that a competent
coder could replicate your experiments, using a different programming language and operating
167 Compare to Footnote 3, p. 9.
168 “Sun Refines Randomness”: http://alife.co.uk/nonrandom/
195
system, and still get more or less the same results. Otherwise, who’s to know if you just didn’t make
this stuff up? To make replicable experiments you’ll need to describe your algorithm and relevant
parameters in sufficient detail. Pseudocode would be nice.
Even if you have described your algorithm in detail, if the algorithm is gigantic and absurdly
complex, it’s not considered replicable. You can’t just thumb your nose at your readers and say
”replicate this giant monster, I dare you.” Instead, you’ll probably need to provide actual code
somewhere for your readers to access so they don’t have to write it themselves. People are scared
of providing code so others can examine it, mostly because they’re ashamed of their code quality.
Be brave.169
Duplicability If you are performing experiments and making claims, it’s helpful to strive not
just for replicability but for the higher standard of duplicability. Here you’re enabling others to
exactly duplicate your results, ideally in environments other than your particular computer. The
difference between replicability and duplicability is fundamental when dealing with a stochastic
system: replicable experiments can be more or less repeated, with results which are statistically
equivalent. Duplicable experiments are exactly the same when run elsewhere. For example, a good
metaheuristics toolkit should be able to enable you to move to a new operating system and a new
CPU and repeat the identical experiment.170 To get duplicability, you’ll need to think about your
language and environment choice.
Why is this important? Let’s say you’ve published some experiments, and Person X approaches
you telling you he can’t replicate your results. Uh oh. “No problem”, you say, and you hand him
your code. Then he tries to run the code on his system and gets... a different result. How do you
prove your claims are still valid? Could it be a bug in his operating system, compiler, or CPU? Or
yours? Did you forget to give him the specific random number generator seeds that produce the
given result? It’s for these reasons that duplicability provides a bit of piece of mind. Replicability is
crucial; duplicability would be nice. Consider it.
11.1.2 Comparing Techniques

By far the most common kind of experiment you’ll find yourself doing in metaheuristics is compar-
ing two different techniques. For example, let’s say you want to show that, on some problem Foo,
if you apply Particle Swarm Optimization with α = 0.9, β = 0.1, γ = 0.1, δ = 0, % = 1, and with a
population of size 10, you’ll get better results than if you use the 5 + 1 Evolution Strategy using
Gaussian Convolution with σ2 = 0.1. How do you do this?
By What Yardstick Should We Compare our Techniques? This is the first question that needs
to be answered. At the end of a run, you often are left with a single best solution (or at least one
169 Imust admit, I often am not. But I try to be.
170 And now we come to the delicate point where I suggest that you may wish to consider a language other than C++:
it’s not a language which makes duplicability easy. C++ and C depend critically on the specifics of your CPU: how
large is a long? How is cos performed? How about sqrt? Is your CPU big-endian, little-endian, or something else?
Does compiling with certain floating-point optimizations turned on change the results? It can be frustrating to get
results running on Machine A, only to recompile on Machine B and get something subtly, but importantly, different.
Perhaps with everyone using the same Intel processors these days, it’s less of a concern. But still, consider picking a
“safe” language: Java in particular can provide precise duplicable results if you need it to.
196
which isn’t worse than any of the others). The quality or fitness of this solution is known as the
best of run. In most cases you’d like this best of run quality to be as good as possible.
For most metaheuristics comparisons your goal is to demonstrate that technique A in some
sense performs better than technique B with regard to best of run quality. Nowadays evaluations
are the primary cost in in metaheuristics, so most researchers tend to ask the following question: if
you could do a single run with a fixed budget of m evaluations, and needed a solution of the highest
quality possible, which technique should you pick? This is exactly the same thing as asking: which
technique has the highest expected (or mean) best of run?171
An alternative question that has been asked before is: how many evaluations do I need to run
before I reach some level L of quality? Often L is simply defined as “the optimum”. Or: if I run
my technique n times, how often do I reach this level? Such formulations have taken many guises
in the past, but the most common one, found in the genetic programming world, is the so-called
computational effort measure.
It is my opinion that this alternative question usually isn’t a good question to ask. Metaheuristics
are applied to hard problems. If you’re gauging techniques by how quickly they solve a problem,
then your problem is trivial and your claims are may be unhelpful for more realistic problems.
Further, such measures are somewhat challenging to establish statistical significance for, and
computational effort in particular may be less accurate than hoped for.172
A third question comes from the machine learning community: if I find a candidate solution
which does well for some set T of test cases, how well is this solution likely to perform in the real
world? This is a question of generalizability: we’re asking how well technique A learns about the
world from a small sample (T) of inputs. One simple approach to gauging this is to create two
disjoint sets of test cases T and S. You can make T however large you like, but I’d make S relatively
large, perhaps 100. T will be the test cases used to to develop our solution (commonly called the
training set). Once we have a final solution, we gauge its quality by applying it to the test cases in
S — which it has never seen before — and seeing how well it performs. S is called the test set. There
exist more nuanced methods for doing train/test methodologies, such as k-fold cross validation,
but the one described is very common.
Finally, multiobjective problems pose special difficulties, because the result of a multiobjective
run is not a single solution but a whole set of solutions which lie along the Pareto front. As a
result, there really is no satisfactory way to compare multiobjective optimization techniques. Still though,
171 What if you could run a technique five times and take the best result of the five? Which is better then? It turns out,
it’s not necessarily A. If A had a mean of 5 but a variance of 0.01, while B had a mean of 4 (worse) but a variance of 20,
you’d pick A if you ran just once, but you’d prefer B if you could run more than once and take the maximum of the runs.
172 Liviu Panait and I wrote a paper attacking the philosophy behind computational effort and similar measures and
noting its poor correlation with expected-quality measures: Sean Luke and Liviu Panait, 2002, Is the perfect the enemy of
the good?, in W. B. Langdon, et al., editors, GECCO 2002: Proceedings of the Genetic and Evolutionary Computation Conference,
pages 820–828, Morgan Kaufmann Publishers, New York.
Steffan Christensen and Franz Oppacher have also been tough on the computational effort measure: they’ve established
that it significantly underestimates the true effort: Steffen Christensen and Franz Oppacher, 2002, An analysis of Koza’s
computational effort statistic for genetic programming, in James A. Foster, et al., editors, Genetic Programming, Proceedings
of the 5th European Conference, EuroGP 2002, volume 2278 of LNCS, pages 182–191, Springer-Verlag.
Matthew Walker, Howard Edwards, and Chris Messom been establishing methods to compute statistical significance
for the computational effort measure. If you’re interested in going after the alternative question, you should definitely
try to use a method like theirs to add some rigor to any claims. Their latest work is Matthew Walker, Howard Edwards,
and Chris Messom, 2007, The reliability of confidence intervals for computational effort comparisons, in Dirk Thierens,
et al., editors, GECCO ’07: Proceedings of the 9th Annual Conference on Genetic and Evolutionary Computation, volume 2,
pages 1716–1723, ACM Press.
197
researchers have to do something. Eckart Zitzler, Kalyanmoy Deb, and Lothar Thiele proposed
various measures for comparing techniques173 which are still in wide use today. Many of these
techniques assume that you know beforehand what the true Pareto front is: this probably will not
be true for real problems. Much research is now turning towards comparing techniques based on
which has the largest hypervolume — the volume of the multiobjective space dominated by the
front discovered by the technique. Hypervolume is, unfortunately, nontrivial and expensive to
compute.
Statistical Significance Okay so you’ve settled on a question to ask and a way of getting results
out of your Particle Swarm Optimization and Evolution Strategy techniques. You run PSO once
and get a 10.5. You run your Evolution Strategy once and get a 10.2. So PSO did better, right?
Nope. How do you know that your results aren’t due to the random numbers you happened to
get from your generator? What happens if you run a second time with a different random number
generator seed? Will PSO still beat ES then or will it be the other way around? Keep in mind that
this is a stochastic technique, not a deterministic one. To determine that PSO really is better than ES
for problem Foo, you’ll need to run some n times and take the average. To eliminate the possibility
of randomness messing with our results, n needs to be large.
You could do this trivially by running your problems A and B, say, a billion times each, and
comparing their means. But who has time to do a billion runs? We need a way to state with
some definiteness that A is better than B after testing A and B each some smaller number of times:
perhaps 50 or 100. To do this, we need a hypothesis test.
The literature on hypothesis tests is huge, and there are many options. Here my goal is to suggest
a couple of approaches which I think will serve you well for the large majority of situations you
may find yourself in. Before we get to hypothesis tests, let’s begin with some strong suggestions:
• Unless you know what you’re doing, always run each technique at least 30 times. I strongly
suggest 50 or 100 times per technique. The more runs you do, the easier it is to prove that the
techniques produce different expected results.
• Each run should be independent — there should be no relationship between the runs. In
particular, each run should employ a unique random number seed.
• Be as conservative as you possibly can with regard to your claim. Don’t just compare
your newfangled Particle Swarm method against a specific Evolution Strategy. Instead, try
Evolution Strategies with lots of different parameter settings to find the one which performs
the best. Compare your new method against that best-performing one. Make it as hard as
possible for your claim to succeed.
Okay, so you’ve done all these things. You now have 100 independent results for technique A
and 100 independent results for technique B. The mean of the A results is better (let’s say, higher)
than the mean of the B results. What do you do now?
Your hypothesis is that A is better than B. The null hypothesis — your enemy — claims that
there’s no difference between the two, that is, the perceived difference is just due to your random
numbers. You need to compute what the probability is that the null hypothesis is wrong. You want
173 EckartZitzler, Kalyanmoy Deb, and Lothar Thiele, 2000, Comparison of multiobjective evolutionary algorithms:
Empirical results, Evolutionary Computation, 8(2), 125–148
198
Desired Probability Desired Probability Desired Probability
dof 95% 98% 99% 99.8% dof 95% 98% 99% 99.8% dof 95% 98% 99% 99.8%
1 12.706 31.821 63.657 318.313 35 2.030 2.438 2.724 3.340 69 1.995 2.382 2.649 3.213
2 4.303 6.965 9.925 22.327 36 2.028 2.434 2.719 3.333 70 1.994 2.381 2.648 3.211
3 3.182 4.541 5.841 10.215 37 2.026 2.431 2.715 3.326 71 1.994 2.380 2.647 3.209
4 2.776 3.747 4.604 7.173 38 2.024 2.429 2.712 3.319 72 1.993 2.379 2.646 3.207
5 2.571 3.365 4.032 5.893 39 2.023 2.426 2.708 3.313 73 1.993 2.379 2.645 3.206
6 2.447 3.143 3.707 5.208 40 2.021 2.423 2.704 3.307 74 1.993 2.378 2.644 3.204
7 2.365 2.998 3.499 4.782 41 2.020 2.421 2.701 3.301 75 1.992 2.377 2.643 3.202
8 2.306 2.896 3.355 4.499 42 2.018 2.418 2.698 3.296 76 1.992 2.376 2.642 3.201
9 2.262 2.821 3.250 4.296 43 2.017 2.416 2.695 3.291 77 1.991 2.376 2.641 3.199
10 2.228 2.764 3.169 4.143 44 2.015 2.414 2.692 3.286 78 1.991 2.375 2.640 3.198
11 2.201 2.718 3.106 4.024 45 2.014 2.412 2.690 3.281 79 1.990 2.374 2.640 3.197
12 2.179 2.681 3.055 3.929 46 2.013 2.410 2.687 3.277 80 1.990 2.374 2.639 3.195
13 2.160 2.650 3.012 3.852 47 2.012 2.408 2.685 3.273 81 1.990 2.373 2.638 3.194
14 2.145 2.624 2.977 3.787 48 2.011 2.407 2.682 3.269 82 1.989 2.373 2.637 3.193
15 2.131 2.602 2.947 3.733 49 2.010 2.405 2.680 3.265 83 1.989 2.372 2.636 3.191
16 2.120 2.583 2.921 3.686 50 2.009 2.403 2.678 3.261 84 1.989 2.372 2.636 3.190
17 2.110 2.567 2.898 3.646 51 2.008 2.402 2.676 3.258 85 1.988 2.371 2.635 3.189
18 2.101 2.552 2.878 3.610 52 2.007 2.400 2.674 3.255 86 1.988 2.370 2.634 3.188
19 2.093 2.539 2.861 3.579 53 2.006 2.399 2.672 3.251 87 1.988 2.370 2.634 3.187
20 2.086 2.528 2.845 3.552 54 2.005 2.397 2.670 3.248 88 1.987 2.369 2.633 3.185
21 2.080 2.518 2.831 3.527 55 2.004 2.396 2.668 3.245 89 1.987 2.369 2.632 3.184
22 2.074 2.508 2.819 3.505 56 2.003 2.395 2.667 3.242 90 1.987 2.368 2.632 3.183
23 2.069 2.500 2.807 3.485 57 2.002 2.394 2.665 3.239 91 1.986 2.368 2.631 3.182
24 2.064 2.492 2.797 3.467 58 2.002 2.392 2.663 3.237 92 1.986 2.368 2.630 3.181
25 2.060 2.485 2.787 3.450 59 2.001 2.391 2.662 3.234 93 1.986 2.367 2.630 3.180
26 2.056 2.479 2.779 3.435 60 2.000 2.390 2.660 3.232 94 1.986 2.367 2.629 3.179
27 2.052 2.473 2.771 3.421 61 2.000 2.389 2.659 3.229 95 1.985 2.366 2.629 3.178
28 2.048 2.467 2.763 3.408 62 1.999 2.388 2.657 3.227 96 1.985 2.366 2.628 3.177
29 2.045 2.462 2.756 3.396 63 1.998 2.387 2.656 3.225 97 1.985 2.365 2.627 3.176
30 2.042 2.457 2.750 3.385 64 1.998 2.386 2.655 3.223 98 1.984 2.365 2.627 3.175
31 2.040 2.453 2.744 3.375 65 1.997 2.385 2.654 3.220 99 1.984 2.365 2.626 3.175
32 2.037 2.449 2.738 3.365 66 1.997 2.384 2.652 3.218 100 1.984 2.364 2.626 3.174
33 2.035 2.445 2.733 3.356 67 1.996 2.383 2.651 3.216 ∞ 1.960 2.326 2.576 3.090
34 2.032 2.441 2.728 3.348 68 1.995 2.382 2.650 3.214
Table 4 Table of t-values by degrees of freedom (dof ) and desired probability that the Null Hypothesis is wrong (2-tailed
t-tests only). To verify that the Null Hypothesis is wrong with the given probability, you need to have a t-value larger
than the given value. If your degrees of freedom exceed 100, be conservative: use 100, unless they’re huge, and so you
can justifiably use ∞. 95% is generally an acceptable minimum probability, but higher probabilities are preferred.
that probability to be as high as possible. To be accepted in the research community, you usually
need to achieve at least a 95% probability; and ideally a 99% or better probability.
A hypothesis test estimates this probability for you. Hypothesis tests come in various flavors:
some more often claim that A is better than B when in fact there’s no difference. Others will more
conservatively claim that there’s no difference between A and B when in fact there is a difference.
You always want to err on the side of conservatism.
199
The most common hypothesis test, mostly because it’s easy to do, is Student’s t-Test.174 The
most conservative such t-Test is one which doesn’t presume that the results of A and B come from
distributions with the same variance. We’ll use the “two-tailed” version of the test. To do this,
you first need to compute the means µ A , µ B , variances σA2 , σB2 , and number of results (n A , n B , in our
example, n A = n B = 100) for technique A and technique B respectively. With these you determine
the t statistic and the degrees of freedom.
µ A − µB
t= ' 2
σA σB2
n A + nB
6 2
72
σA σB2
nA + nB
degrees of freedom = 6 2
72 6 72
σA σB2
nA / ( n A − 1) + nB / ( n B − 1)
Let’s say your degrees of freedom came out to 100 and you have chosen 95% as your probability.
From Table 4, we find that you must have a t value of 1.984 or greater. Imagine that that your
t value came out as, oh, let’s say, 0.523. This tells us that you have failed to disprove the Null
Hypothesis with an adequate probability. Thus you have no evidence that PSO is actually better
than ES for the Foo problem.
As you can see from the table, if you want to make it easier to pass the t-test, the way to do it
is to increase your degrees of freedom. This translates into doing more runs (that is, increasing
n A and n B ). More runs is always good! But beware: if you need a very large number of runs to
do this, it’s likely the case that though your techniques are different, the difference is very small.
Now you’ll run up against the “so what?” question: so what if PSO eeks out just barely better
results than ES on problem Foo? Thus what you usually want to be able to argue is both (1) that
the difference between your two techniques statistically significant, that is, that a hypothesis test
agrees with you that it actually exists; and (2) that the difference is also considerable and likely to
be important.
The t-Test should be viewed as the absolute minimum you should do for published work.
Anything less and you should be ashamed of yourself. The problem with the t-Test — and it’s a big
problem — is that it is parametric, that is, it relies solely on the mean, variance, and sample count of
your results. This is because the t-Test makes a huge assumption: that the results produced by your
techniques A and B are each drawn from a normal (Gaussian) distribution.
In metaheuristics scenarios, that’s almost never true.
A great many metaheuristics problems produce results which are fairly skewed. Now the t-Test
is pretty robust even with relatively skewed data. But if the data is too skewed, the t-Test starts
being less conservative than it should. Also very bad for the t-Test is data with multiple peaks.
To compensate for this, there’s a better approach: a nonparametric hypothesis test. This kind of
test ignores the actual values of your data and only considers their rank ordering with respect to one
another.175 As a result, such tests are much less sensitive, but they are not fooled by assumptions
about how your results are distributed. If you pass a non-parametric test, few can criticize you.
174 It’s called this because it’s based on work by William Sealy Gosset around 1908, who worked at Guinness Brewery
and secretly published under the pseudonym “Student”. He did so because Guinness wouldn’t allow its workers to
publish anything out of fear of leaking trade secrets. The t-Test itself was, however, mostly derived by Ronald Aylmer
Fisher, a famous statistician who conversed with Gosset and made his work popular.
175 Sound familiar? Think: fitness-proportionate selection versus tournament selection.
200
There are a various nonparametric tests, but Mark Wineberg and Steffen Christensen176 suggest
a particularly simple and effective one:
1. Throw all the results of techniques A and B together into one vector.
2. Sort the vector by result value.
3. Replace the result values with their rank values (that is, their locations in the vector).
4. Results with the same value are assigned the average of their combined ranks.
5. Break the results back into the technique-A results and the technique-B results.
6. Using the rank values rather than the original result values, do a t-Test.
Let’s do an example. Imagine that, against good judgement and the recommendations of this
text, you have decided only to do five runs of each technique (PSO and ES). Your results were:
PSO: 0.1 0.5 0.8 0.9 0.9 ES: 0.2 0.3 0.5 0.7 0.9
We put them together into one 0.1 0.2 0.3 0.5 0.5 0.7 0.8 0.9 0.9 0.9
vector and sort it. PSO ES ES ES PSO ES PSO ES PSO PSO
1 2 3 4 5 6 7 8 9 10
Next we include ranks. 0.1 0.2 0.3 0.5 0.5 0.7 0.8 0.9 0.9 0.9
PSO ES ES ES PSO ES PSO ES PSO PSO
1 2 3 4.5 4.5 6 7 9 9 9
Next we average ranks for
0.1 0.2 0.3 0.5 0.5 0.7 0.8 0.9 0.9 0.9
results with the same values.
PSO ES ES ES PSO ES PSO ES PSO PSO
Next we replace the values 1 2 3 4.5 4.5 6 7 9 9 9

with just the ranks. PSO ES ES ES PSO ES PSO ES PSO PSO
Finally, we break the results back out into their groups again. The ranks are all that are left.
PSO: 1 4.5 7 9 9 ES: 2 3 4.5 6 9
We can now do a plain-old t-Test on these revised values instead. Note that we’re no longer
testing whether the means of the two techniques are different from one another. Instead, since we’re
looking at rank orderings, it’s somewhat closer to saying that the medians of the two techniques
differ. It’s still a better measure than a plain t-Test by a long shot.
176 Seethe very last entry in Section 11.3.1 for pointers to their excellent lecture slides. A number of suggestions here
were inspired from those slides.
201
Comparing More than Two Techniques t-Tests only compare two techniques. Let’s say you have
five techniques, A, B, C, D, and E. You want to prove that A does better than the rest. How do you
compare them? One approach is to compare A against B (with a hypothesis test), then A against C,
then A against D, then A against E. If you do this, remember that it’s critical that each time you
compare A against another technique, you should do a new set of independent runs for A, with
new random number generator seeds. Don’t reuse your old runs. Or perhaps you want to compare
each method against every other method: that is, A versus B, A versus C, A versus D, A versus E,
B versus C, B versus D, B versus E, C versus D, C versus E, and finally D versus E. Phew! Again,
remember that each comparison should use new, independent runs.
Doing individual pairwise hypothesis tests isn’t sufficient though. Keep in mind that the
point of a hypothesis test is to compute the probability that your claim is valid. If you do a single
comparison (A versus B) at 95% probability, there is a 5% chance that your claim is false. But if
you compare A against four other techniques (A versus B, A versus C, A versus D, A versus E),
each at 95% probability, you have an approximately 20% chance that one of them is false. If you
compared each method against the others, resulting in ten comparisons, you have an approximately
50% chance that one of them is false! It’s pretty common that you’ll do a lot of experiments in your
paper. And so with a high probability one of your hypothesis tests will come up false.
It’s better style to try to fix this probability, and ideally get it back up to 95% (or whatever value
you had originally chosen). The simplest way to do this is to apply the Bonferroni correction.
Specifically, if you have m comparisons to do, and the desired probability of one of them being
wrong is p total, then revise each individual probability of being wrong to be p/m, and thus the
probability of being right is 1 − p/m. In our examples above, if we wish to compare A against the
other techniques (four comparisons), and want to retain a 95% probability of being right — that is,
a 1/20 chance of being wrong, then each of our comparisons should be done with a 1/20 4 = 1/80
probability of being wrong. That translates into using a 1 − 1/80 = 0.9875% probability for each
hypothesis test. Similarly, if you’re comparing all the techniques (ten comparisons), you’ll have
1 − 1/200 = 0.995%. Not easy to beat!
A much less extreme method, in terms of how high your probability has to go, is the ANOVA,
a fairly complex method which compares m techniques at one time and tells you if any one of them is
different from the others. Interestingly, the ANOVA doesn’t tell you which techniques are different
from which others: for that you apply a so-called post-hoc comparison, the most conservative of
which (always be conservative!) is the Tukey comparison.177 One difficulty with the ANOVA is
that, like the original t-Test, it assumes that your distributions are normal. Which is rarely the case.
There exist non-parametric ANOVA methods as well. The ANOVA (and related tests) are far too
complex to describe here: consult a good statistics book.
One of the strange effects you’ll get when comparing m techniques is nontransitivity among
your results. For example, let’s say that, looking at their means, A > B > C > D > E. But when
you run the ANOVA, it tells you that A and B aren’t statistically different, and B and C aren’t
statistically different, but A and C are statistically significantly different! Furthermore, D and
E aren’t statistically different, but A, B, and C are all statistically significantly different from D
and E. Eesh. How do you report something like this? Usually, with overbars connecting groups
with no significant difference among them: A B C D E Be sure to notice the overlapping but
unconnected overbars over A, B, and C.
177 Named after the statistician John Tukey.
202
11.2 Simple Test Problems
The test problems below are common, and sometimes trivial, fitness or quality functions suitable
for small experiments and projects. Problems are provided for fixed-length boolean and real-valued
vectors, multiobjective scenarios, and Genetic Programming (and Grammatical Evolution).
Many of these problems have been overused in the field and are a bit dated: if you’re working
on a scientific research paper, you ought to spend some time examining the current benchmarks
applied to techniques like yours. Also: if you’re using test problems as benchmarks to compare
techniques, be wary of the temptation to shop for benchmarks, that is, to hunt for that narrow set
of benchmark problems that happens to make your technique look good. You can always find
one, but what have you gained? Instead, try to understand how your technique performs on a
wide range of well-understood problems from the literature, or on problems of strong interest to a
specific community.178
11.2.1 Boolean Vector Problems

Max Ones Max Ones, sometimes called OneMax, is a trivial boolean problem: it’s the total
number of ones in your vector. This is the classic example of a linear problem, where there is no
linkage between any of the vector values at all. Simple Hill-Climbing can solve this problem easily.
Max Ones is due to David Ackley:179
n
f ($ x1 , ..., xn %) = ∑ xi
i =1
Leading Ones This problem is also quite simple: it counts the number of ones in your vector,
starting at the beginning, until a zero is encountered. Put another way, it returns the position of the
first zero found in your vector (minus one). The equation below is a clever way of describing this
mathwise, but you wouldn’t implement it like that — too expensive. Just count the ones up to the
first zero. Leading Ones is not a linear problem: the contribution of a slot xi in the vector depends
critically on the values of the slots x1 , ..., xi−1 . Nonetheless, it’s pretty simple to solve.
n i
f ($ x1 , ..., xn %) = ∑ ∏ xj
i =1 j =1
Leading Ones Blocks This variant of Leading Ones is somewhat more challenging. Given
a value b, we count the number of strings of ones, each b long, until we see a zero. For ex-
ample, if b = 3, then f ($1, 1, 0, 0, 0, 1, 1, 0, 1%) = 0 because we don’t have a string of 3 at the
178 At this point it’s worth bringing up the infamous No Free Lunch Theorem, or NFL, by David Wolpert and William
Macready. The NFL stated that within certain constraints, over the space of all possible problems, every optimization
technique will perform as well as every other one on average (including Random Search). That is, if there exists a set
of problems P for which technique A beats technique B by a certain amount, there also exists an equal-sized set of
problems P) for which the opposite is true. This is of considerable theoretical interest but, I think, of limited practical
value, because the space of all possible problems likely includes many extremely unusual and pathological problems
which are rarely if ever seen in practice. In my opinion, of more of interest is what kinds of techniques perform well on
the typical problems faced by practitioners, and why. For more on the NFL, see David Wolpert and William Macready,
1997, No free lunch theorems for optimization, IEEE Transactions on Evolutionary Computation, 1(1), 67–82.
179 David Ackley, 1987, A Connectionist Machine for Genetic Hillclimbing, Kluwer Academic Publishers.
203
beginning yet. But f ($1, 1, 1, 0, 0, 0, 1, 0, 1%) = 1. Furthermore, f ($1, 1, 1, 1, 0, 1, 1, 0, 1%) = 1 but
f ($1, 1, 1, 1, 1, 1, 0, 1, 0%) = 2, and ultimately f ($1, 1, 1, 1, 1, 1, 1, 1, 1%) = 3. A simple way to do this is
to do Leading Ones, then divide the result by b, and floor it to the nearest integer:
8 9
1 n i
b i∑ ∏ xj
f ($ x1 , ..., xn %) =
=1 j =1
Trap The so-called Trap Problems are classic examples of deceptive functions.. Here’s a simple
one which is easily described: the fitness of your vector is the number of zeros in the vector, unless
you have no zeros at all, in which case the fitness of the vector is the suddenly optimally high (n + 1).
Thus this problem sets up a gradient to lead you gently away from the optimal all-ones (no zeros)
case, and deep into the trap. For example, f ($0, 0, 0, 0%) = 4, f ($0, 0, 1, 0%) = 3, f ($1, 0, 1, 0%) = 2,
f ($1, 0, 1, 1%) = 1, but boom, f ($1, 1, 1, 1%) = 5. A clever math formulation of this has two terms:
the sum part is the number of zeros in the vector. The product part only comes into play when you
have all ones. Various trap functions were originally due to David Ackley.180
: ;
n n
f ($ x1 , ..., xn %) = n − ∑ xi + ( n + 1) ∏ x i
i =1 i =1
11.2.2 Real-Valued Vector Problems

Many classic real-valued vector problems are minimization problems rather than maximization
ones. To convert them to a maximization problem, the simplest solution is to negate the result. If
you’re using Fitness Proportionate Selection or SUS, you’ll also need to add a big enough number
that there aren’t any negative values. I’d use Tournament Selection instead.
Most of the problems described below are shown, in the trivial 2-dimensional case, in Figure 67.
Sum Sum is the trivial real-valued version of Max Ones. It’s just the sum of your vector. As
would be expected, Sum is a linear problem and so has no linkage.
n
f ($ x1 , ..., xn %) = ∑ xi xi ∈ [0.0, 1.0]
i =1
Linear Linear functions are the generalization of Sum, and again have no linkage at all. They’re
just the weighted sum of your vector, where each weight is given by a constant ai . Given a vector
of constants $ a0 , ..., an %, which you provide, we weight each element, then add them up:
n
f ($ x1 , ..., xn %) = a0 + ∑ ai xi xi ∈ [0.0, 1.0]
i =1
180 As was Max Ones. See Footnote 179, p. 203.
204
-5 -5 -5
5 -2.5 5 -2.5 5 -2.5 -2
0 2 -1
0 0
2.5 2.5 2.5 2.5 2.5 2.5 0
1
5 5 5 1
0 0 0
0 2
-2.5 10 -2.5 20 -2.5
60 -1
-5 -5 -5 2000
15
-2
40
0 10
1000
5 20
-10
0 0 0
Sum Step Sphere Rosenbrock

-40
-5 -400 -500 40 -20
5 -2.5 400 -200 500 -250 0
0 0 0 20 20
2.5 2.5 200 200 250 250 40
5 400 500 0
0 0 0
-20 3
-2.5 1500 -200 -250
500 150 -40
-5 -400 -500
1000 2
0 100
500
50 1
0 -500
0 0
Rastrigin Schwefel Griewank Griewank (Detail)

Figure 67 Real-valued problems in two dimensions ($ x1 , x2 %).
Step Another no-linkage function, but this time it’s got a wrinkle. Because it uses the floor
function, there are regions where small mutations in any given floating point value don’t change
fitness at all. This function is part of a popular test suite by Ken De Jong,181 and so has traditional
bounds on the xi values (between −5.12 and +5.12 inclusive). The function is usually minimized,
though it doesn’t matter much: you can search for the maximum too, it’s about the same.
n
(Minimize) f ($ x1 , ..., xn %) = 6n + ∑ 0 xi 1 xi ∈ [−5.12, 5.12]
i =1
Sphere Our last no-linkage problem, due to Ingo Rechenberg.182 Here we’re summing the squares
of the individual elements. This is again a minimization problem, and is part of De Jong’s test suite
(note the bounds). Maximization is also interesting, as there are global maxima at the corners.
n
(Minimize) f ($ x1 , ..., xn %) = ∑ xi2 xi ∈ [−5.12, 5.12]
i =1
181 Perhaps too popular. Ken De Jong has been waging a campaign to get people to stop using it! The test suite was
proposed in De Jong’s PhD thesis: Kenneth De Jong, 1975, An Analysis of the Behaviour of a Class of Genetic Adaptive
Systems, Ph.D. thesis, University of Michigan. The thesis is available online at http://cs.gmu.edu/∼eclab/kdj thesis.html
182 Ingo Rechenberg, 1973, Evolutionsstrategie: Optimierung technischer Systeme nach Prinzipien der biologischen Evolution,
Fromman-Holzbook, Stuttgart, Germany.
205
Rosenbrock A classic optimization problem well predating the field, from Howard Rosen-
brock.183 In two dimensions, this function creates a little valley bent around a low hill, with
large wings on each side. The minimum is at $1, 1, ..., 1%, in the valley on one side of the low hill, and
individuals often get stuck on the other side. The traditional bounds are shown. It’s a minimization
problem.
n −1
(Minimize) f ($ x1 , ..., xn %) = ∑ (1 − xi )2 + 100(xi+1 − xi2 )2 xi ∈ [−2.048, 2.048]
i =1
Rastrigin Originally proposed by Leonard Andreevich Rastrigin184 in 1974 as a two-dimensional

function, and later extended by Heinz Mühlenbein, M. Schomisch, and Joachim Born to more
variables.185 This function is essentially a large egg carton bent under a basketball: it’s a combination
of Sphere and a sine wave which creates a great many local optima. It’s a minimization problem.
Some literature has xi ∈ [−5.12, 5.12], following De Jong’s tradition (that’s what I’m doing here),
but others use different bounds.
n
(Minimize) f ($ x1 , ..., xn %) = 10n + ∑ xi2 − 10 cos(2πxi ) xi ∈ [−5.12, 5.12]
i =1
Schwefel This function, due to Hans-Paul Schwefel,186 has many local optima like Rastrigin; but
is organized so that the local optima are close to one another (and thus easier to jump to) the further
you get from the global optima. It’s thus described as a deceptive problem. Again, minimization.
Notice the larger traditional bounds than we’ve seen so far.
n 4' 5
(Minimize) f ($ x1 , ..., xn %) = ∑ − xi sin | xi | xi ∈ [−512.03, 511.97]
i =1
Some variations add 418.9829×n to the function to set the minimum to about 0.
Griewank Not to be outdone by Rastrigin, Andreas Griewank’s similar function has a zillion
local optima.187 The function is minimized, and traditionally has bounds from −600 to +600, which
this creates massive numbers of local optima.
: ; 4 5
n n
1 x
(Minimize) f ($ x1 , ..., xn %) = 1 + ∑ xi2 + ∏ cos √i xi ∈ [−600, 600]
4000 i =1 i =1 i
183 Howard Rosenbrock, 1960, An automatic method for finding the greatest or least value of a function, The Computer
Journal, 3(3), 174–184.
184 I believe this was from Leonard Andreevich Rastrigin, 1974, Systems of Extremal Control, Nauka, in Russian. Nearly
impossible to get ahold of, so don’t bother.

185 Heinz Mühlenbein, D. Schomisch, and Joachim Born, 1991, The parallel genetic algorithm as function optimizer, in
Richard Belew and Lashoon Booker, editors, Proceedings of the Fourth International Conference on Genetic Algorithms, pages
271–278.
186 Hans-Paul Schwefel, 1977, Numerische Optimierung von Computer-Modellen mittels der Evolutionsstrategie, Birkhauser.
187 Andreas Griewank, 1981, Generalized descent for global optimization, Journal of Optimization Theory and Applications,
34, 11–39.
206
1 1
0.8 0.8
0.6 0.6
Objective 2
Objective 2
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Objective 1 Objective 1
ZDT1 ZDT2
1
1.2
0.75
1
0.5
0.8
Objective 2
Objective 2
0.25
0.6
0
-0.25 0.4
-0.5 0.2
-0.75
0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 1
Objective 1 Objective 1
ZDT3 ZDT4
Pareto front is discontinuous. Thin line indicates the highest local suboptimal pareto
front. Other local suboptimal pareto fronts not shown.
Figure 68 Pareto fronts of four multiobjective problems (ZDT1, ZDT2, ZDT3, and ZDT4) as described in Section 11.2.3.
All four problems are minimization problems, so lower objective values are preferred.
11.2.3 Multiobjective Problems

The problems described below are all from a classic multiobjective comparison paper by Eckart
Zitzler, Kalyanmoy Deb, and Lothar Thiele.188 Like many multiobjective test problems, they’re all
set up for minimization: you can change this to maximization by negating (for example). All four
problems have two objectives O1 and O2 . The problems are all designed such that O2 is a function
of two auxillary functions g and h. The global Pareto fronts for all four problems, and in one case a
strong local Pareto front, are all shown in Figure 68.
188 Eckart Zitzler, Kalyanmoy Deb, and Lothar Thiele, 2000, Comparison of multiobjective evolutionary algorithms:
Empirical results, Evolutionary Computation, 8(2), 125–148.
207
ZDT1 This is a basic multiobjective problem with a convex Pareto front for real-valued vector
individuals n = 30 genes long. The problem has no local optima.
(Minimize) O1 ($ x1 , ..., xn %) = x1 xi ∈ [0, 1]
O2 ($ x1 , ..., xn %) = g($ x1 , ..., xn %) × h($ x1 , ..., xn %)
n
9
n − 1 i∑
g($ x1 , ..., xn %) = 1 + xi
=2
<
x1
h($ x1 , ..., xn %) = 1 −
g($ x1 , ..., xn %)
ZDT2 This function is like ZDT1, but is concave. Again, n = 30. The problem has no local optima.
O2 ($ x1 , ..., xn %) = g($ x1 , ..., xn %) × h($ x1 , ..., xn %)
n
9
g($ x1 , ..., xn %) = 1 + ∑
n − 1 i =2
xi
4 52
x1
h($ x1 , ..., xn %) = 1 −
g($ x1 , ..., xn %)
ZDT3 This function has a discontinuous Pareto front. Again, n = 30. The problem has no local
optima.
O2 ($ x1 , ..., xn %) = g($ x1 , ..., xn %) × h($ x1 , ..., xn %)
n
9
n − 1 i∑
g($ x1 , ..., xn %) = 1 + xi
=2
<
x1 x1
h($ x1 , ..., xn %) = 1 − − sin(10πx1 )
g($ x1 , ..., xn %) g($ x1 , ..., xn %)
ZDT4 This function has a convex Pareto front but has a many local suboptimal Pareto fronts to
trap individuals, making this a moderately challenging problem. The problem is defined for a
smaller value of n than the others: n = 10. The value x1 ranges in [0, 1], but the other xi all range in
[−5, 5].
(Minimize) O1 ($ x1 , ..., xn %) = x1 x1 ∈ [0, 1], xi>1 ∈ [−5, 5]
O2 ($ x1 , ..., xn %) = g($ x1 , ..., xn %) × h($ x1 , ..., xn %)
n
g($ x1 , ..., xn %) = 1 + 10(n − 1) + ∑ xi2 − 10 cos(4πxi )
i =2
<
x1
h($ x1 , ..., xn %) = 1 −
g($ x1 , ..., xn %)
208
11.2.4 Genetic Programming Problems
As they’re optimizing small computer programs, genetic programming problems are somewhat
more colorful, and detailed, than the mathematical functions we’ve seen so far. The problems
described here aren’t very complex: they’re often tackled with a population of 1000 or so, run for 51
generations (including the initial generation). The problems described here are from John Koza.189
Symbolic Regression This is the canoni- Function Arity Description

cal example problem for genetic program-
ming, and is perhaps overused. The ob- (+ i j) 2 Returns i + j
jective is to find a mathematical expres- (− i j) 2 Returns i − j
sion which best fits a set of data points ( ∗ i j) 2 Returns i × j
(% i j) 2 If j is 0, returns 1, else returns i/j
of the form $ x, f ( x )% for some unknown
(sin i) 1 Returns sin(i)
(to the optimization algorithm) function f . (cos i) 1 Returns cos(i)
The traditional function to fit is f ( x ) = (exp i) 1 Returns ei
x4 + x3 + x2 + x, though Koza also sug- (rlog i) 1 If j is 0, returns 0, else returns log(|i |)
gested the functions g( x ) = x5 − 2x3 + x x 0 Returns the value of the independent
and h( x ) = x6 − 2x4 + x2 . These functions variable (x).
are shown in Figure 69. ERCs 0 (Optional) Ephemeral random con-
We begin by creating twenty random stants chosen from floating-point val-
ues from -1 to 1 inclusive.
values x1 , ..., x20 , each between -1 and 1,
which will be used throughout the dura- Table 5 Symbolic Regression Function Set
tion of the run. An individual is assessed as
follows. For each of the 20 xi values, we set the leaf-node function x to return the value of xi , then
evaluate the individual’s tree. The return value from the tree will be called, say, yi . The fitness of
the individual is how close those 20 yi matched their expected f ( xi ), usually using simple distance.
That is, the fitness is ∑20i =1 | f ( x i ) − y i | .
Obviously this is a minimization problem. It’s
1
easily converted to maximization with 1+fitness . An 0.8
example ideal solution is: (+ (* x (* (+ x (* x x)) 0.6
x)) (* (+ x (cos (- x x))) x))
0.4
0.2
11-bit Boolean Multiplexer The objective of the
11-bit Boolean Multiplexer problems is to find a 0
boolean function which performs multiplexing over

a 3-bit address. There are three boolean-valued ad- -0.2
dress variables (A0, A1, and A2) and eight corre- -1 -0.5 0 0.5 1
sponding boolean-valued data variables (D0, D1,

Figure 69 The functions f ( x ): ——, g( x ): – – –, and
D2, D3, D4, D5, D6, D7). The 11-bit Boolean Mul- h( x): - - - - as discussed in the Symbolic Regression sec-
tiplexer problem must return the value of the data tion. f (1) = 4.
variable at the address described by the binary val-
ues of A0, A1, and A2. For example, if A2 is false
189 Adapted from John R. Koza, 1992, Genetic Programming: On the Programming of Computers by Means of Natural Selection,
MIT Press and from John R. Koza, 1994, Genetic Programming II: Automatic Discovery of Reusable Programs, MIT Press.
209
and A1 is true and A0 is true, the address is 3 (binary 011), and so the optimal individual would
return the value stored in D3. Since there are eleven boolean variables altogether, there are 2048
permutations of these variables and hence 2048 test cases. A trivial variant, the 6-bit Boolean
Multiplexer, has two address variables (A0 and A1), four data variables (D0, D1, D2, D3), and 64
test cases.
A Multiplexer individual consists of a Function Arity Description
single tree. To assess the fitness of an in-
dividual, for each test case, the data and (and i j) 2 Returns i ∩ j
address variables are set to return that test (or i j) 2 Returns i ∪ j
case’s permutation of boolean values, and (not i) 1 Returns ¬i
the individual’s tree is then evaluated. The (if test then else) 3 If test is true, then then is re-
fitness is the number of test cases for which turned, else else is returned.
a0, a1, and a2 0 Return the values of variables
the individual returned the correct value
A0, A1, and A2 respectively.
for the data variable expected, given the d0, d1, d2, d3, d4, 0 Return the values of variables
current setting of the address variables. d5, d6, and d7 D0, D1, D2, D3, D4, D5, D6,
An example of an ideal 11-bit Boolean and D7 respectively.
Multiplexer solution is:
Table 6 11-bit Boolean Multiplexer Function Set
(if (not a0) (if (not a0) (if (not a1) (if a2 (if a2 d4 d6) d0) (if a2 d6 (if a2 d4 d2))) (if (or a2 a2) (if a1 (or (if (not (if a2 d5 d0))
(and (and d4 d0) (and a2 d5)) (or (and d7 d0) (not a1))) (if (not a1) (if (if d4 d1 d5) d0 d5) (or d6 (or (and (and d4 d0) (or (and
d5 d1) (and d6 d6))) (and d7 (or (if a0 (or a2 a2) d4) (and d1 (and d5 a2)))))))) d5) (if a1 (or d3 (and d7 d0)) (if a0 d1 d0)))) (if
(or a2 a2) (if a1 (if (not a1) (if (and d7 d0) (if a2 d5 d0) (if a2 d6 d3)) (and d7 (or (if a0 a2 (or d1 a1)) (not a1)))) d5) (if a1 (or
(if (not a0) (if a2 d6 (if a2 d4 d2)) (if a1 d3 (or (or d3 (if a1 d3 d1)) (not a2)))) (not a1)) (if a0 d1 d0))))
Even N-Parity The Even N-Parity prob- Function Arity Description

lems are, like 11-bit Boolean Multiplexer,
also boolean problems over some n number (and i j) 2 Returns i ∩ j
of data variables. In the Even N-Parity prob- (or i j) 2 Returns i ∪ j
lems, the objective is to return true if, for the (nand i j) 2 Returns ¬(i ∩ j)
(nor i j) 2 Returns ¬(i ∪ j)
current boolean settings of these variables,
d0, d1, d2, etc. 0 Return the values of variables D0,
there is an even number of variables whose D1, D2, ... respectively. The num-
value is true. There are thus 2n test cases. ber of dx nodes in the function set
Fitness assessment is basically the same as is the number of bits in the particu-
11-bit Boolean Multiplexer. lar Parity problem being run.
Even N-Parity varies in difficulty de-
Table 7 Even N-Parity Function Set
pending on N, due to the number of test
cases. Bill Langdon notes that Parity doesn’t
have any building blocks.190 An ideal Even
4-Parity solution:
(nand (or (or (nor d3 d0) (nand (or d3 d1) (nor d2 d3))) d3) (nor (nor (and (or (and (or (or (nor d1 d2) (and d3 d0)) (and d1 d2))
(nand (and d0 d3) (nand (or d0 d1) (or d2 d1)))) (and (or d0 d2) (and d1 d1))) (nand (and (nor d3 d0) (and (and (nand (nand (nor
d3 d3) (or (or d0 d0) (nor (and d3 d0) (nor d1 (nand d3 d2))))) d2) (nor d1 d1)) (or (or d0 d1) (nor d3 d2)))) (nand (or d0 d1)
(nor d3 d3)))) (or (and (nand d1 d1) (and d1 d3)) (nor (nand (or d1 d2) (nor d3 d0)) d0))) (and (or (or (or (and (nand d1 d1) (and
d1 d3)) (nor (nand (or d1 d2) (nor d3 d0)) (and (nand d1 d3) (and d3 d0)))) (and d3 d0)) (and d3 d2)) (and (and d1 d2) (or (or
d0 (nor (or d0 d0) (and d2 d3))) d0)))))
190 William Langdon, 1999, Scaling of program tree fitness spaces, Evolutionary Computation, 7(4), 399–428.
210
Artificial Ant Artificial Ant is an oddly Function Arity Description
challenging problem191 for genetic pro-
gramming. The Artificial Ant problem at- (progn3 a b c) 3 a, b, then c are executed.
tempts to find a simple robotic ant algo- (progn2 a b) 2 a, then b are executed.
rithm which will find and eat the most food (if-food-ahead then else) 2 If food is immediately in
pellets within 400 time steps. 192 The ant front of the ant, then is exe-
cuted, else else is executed.
may move forward, turn left, and turn right.
move 0 Moves the ant forward one
If when moving forward it chances across square, eating food if it is
a pellet, it eats it. The ant can also sense there.
if there is a pellet in the square directly in left 0 Rotates the ant ninety de-
front of it. The grid world in which the Ar- grees to the left.
tificial Ant lives is shown in Figure 70. The right 0 Rotates the ant ninety de-
pellet trail shown is known as the “Santa Fe grees to the right.
Trail”. The world is toroidal: walking off an Table 8 Artificial Ant Function Set
edge moves the ant to the opposite edge.
An Artificial Ant individual consists of a single tree.
Fitness assessment works as follows. The ant starts on the
upper-left corner cell, and facing right. The tree is executed:
as each sensory or movement node is executed, the Ant Start Here, Oriented to the Right
senses or moves as told. When the tree has completed ex-
ecution, it is re-executed again and again. Each movement
counts as one time step. Assessment finishes when the
Ant has eaten all the pellets in the world or when the 400
time steps have expired. The Ant’s fitness is the number of
pellets it ate.
The Artificial Ant problem is different from the Sym-
bolic Regression and the boolean problems in that the re-
turn value of each tree node is ignored. The only thing
that matters is each node’s action in the world, that is, each
node’s side effect: moving the ant, turning it, etc. This
means that in Artificial Ant, the order in which the nodes
are executed determines the operation of the individual, Figure 70 The Santa Fe Trail, a toroidal grid
whereas in the previous problems, it doesn’t matter in what world. Black squares indicate pellet locations.
order subtrees are evaluated. A (highly parsimonious) ex-
ample of an optimal Artificial Ant solution is: (progn3 (if-
food-ahead move (progn2 left (progn2 (progn3 right right
right) (if-food-ahead move right)))) move right).
191 One of my all-time favorite papers, mostly due to its Knuth-like excessive attention to detail, is exactly on this topic:
W. B. Langdon and R. Poli, 1998, Why ants are hard, in John R. Koza, et al., editors, Genetic Programming 1998: Proceedings
of the Third Annual Conference, pages 193–201, Morgan Kaufmann.
192 400 may be due to a misprint that has since established itself. John Koza is believed to have actually used 600.
211
Lawnmower In the Lawnmower problem, Function Arity Description
the individual directs a lawnmower to mow (progn2 a b) 2 a, then b are executed. Returns the
a toroidal grid lawn, much as the Artificial return value of b.
Ant domain directs an ant to move about (v8a i j) 2 Evaluates i and j, adds the vectors
its toroidal grid world. In the Lawnmower they return, modulo 8, and returns
domain, an individual may turn left, mow the result.
(frog i) 1 Evaluates i. Let $ x, y% be i’s re-
forwards, or “hop” some $ x, y% units away.
turn value. Then frog moves $ x, y%
Lawnmower has no sensor information: it squares relative to its present rotation,
must be hard-coded to mow the lawn blind. where the positive X axis points in
The standard lawn size is 8 by 8. the present “forward” direction of the
Koza proposed this domain originally lawnmower, and the positive Y axis
to demonstrate the advantages of automat- points in the present “heading left” di-
ically defined functions (ADFs). 193 Lawn- rection. Returns $ x, y%.
mower is difficult without ADFs but fairly mow 0 Moves the lawnmower forward one
square, mowing that square of lawn if
trivial when using ADFs. When not using
it is not already mown. Returns $0, 0%.
ADFs, a Lawnmower individual consists of left 0 Rotates the lawnmower ninety de-
a single tree, and the function set is shown grees to the left. Returns $0, 0%.
in Table 9. When using ADFs, a Lawn- ERCs 0 Ephemeral random constants of the
mower individual consists of three trees: form $ x, y%, where x is an integer cho-
the main tree, an ADF1 tree and an ADF2 sen from the range (0, ..., xmax − 1)
tree; and the function set is augmented as and y is an integer chosen from the
range (0, ..., ymax − 1), where xmax and
described in Table 10.
ymax are the width and height of the
To assess fitness, the lawnmower is lawn in squares, respectively.
placed somewhere on the lawn, and the in-
dividual’s tree is executed once. Each mow Table 9 Lawnmower Function Set.
and frog command moves the lawnmower
and mows the lawn in its new location. Additional ADF functions for Main Tree
Once the tree has been executed, the fitness Function Arity Description
is the number of squares of lawn mown. An (adf1 arg1 ) 1 Automatically defined function
example optimal individual with ADFs: which calls the ADF2 tree.
Main Tree: (progn2 (progn2 (adf1 (progn2 (adf1 left) (v8a adf2 0 Automatically defined function
$7,0% $0,4%))) (progn2 left $3,4%)) (v8a (progn2 (adf1 (v8a which calls the ADF2 tree.
left left)) (progn2 (frog mow) (adf1 adf2))) (adf1 (progn2
(v8a $6,7% adf2) (progn2 $1,1% mow)))))
Additional ADF functions for ADF1 Tree
ADF1: (v8a (v8a (v8a (progn2 (v8a adf2 mow) (v8a Function Arity Description
adf2 mow)) (frog (v8a mow arg1))) (v8a (v8a (frog
arg1) (progn2 $1,4% $2,6%)) (progn2 (v8a $1,5% adf2) (frog adf2 0 Automatically defined function
mow)))) (v8a (v8a (v8a (progn2 adf2 adf2) (v8a adf2 which calls the ADF2 tree.
mow)) (v8a (progn2 arg1 adf2) (frog left))) (frog (v8a (v8a
arg1 left) (v8a $7,0% mow)))))
arg1 0 The value of argument arg1 passed
when the ADF2 tree is called.
ADF2: (progn2 (v8a (progn2 (v8a (v8a mow mow) (v8a
mow $5,1%)) (v8a (v8a mow left) (progn2 left mow))) Removed ADF functions for ADF2 Tree
(v8a (progn2 (v8a mow mow) (progn2 $1,3% $2,1%)) (v8a
(progn2 $3,6% mow) (progn2 left $3,4%)))) (v8a (progn2 Function
(progn2 (v8a mow left) (progn2 $6,6% $1,4%)) (progn2 (v8a (frog i) Removed from the ADF2 tree.
mow left) (v8a mow $7,7%))) (progn2 (v8a (progn2 left
left) (v8a mow left)) (v8a (progn2 left $2,1%) (v8a $1,7%
Table 10 Additions to the Lawnmower Function Set when set
mow)))))
up with two additional ADF trees (ADF1 and ADF2). All three
193 I’ve reordered/renamed Koza’s original ADFs. trees have the same function set except where noted above.
212
Although this individual looks imposing, in fact with ADFs Lawnmower is fairly easy for
genetic programming to solve. Much of this individual is junk. The reason ADFs work so much
better in this domain is simple and unfair: a Lawnmower individual is executed only once, and has
no iteration or recursion, and so within its tree must exist enough commands to move lawnmower
to every spot of lawn. To do this for a single tree demands a big tree. But with when using ADF
trees, the main tree can repeatedly call ADFs (and ADF1 can repeatedly call ADF2), so the total size
of the individual can be much smaller and still take advantage of many more total moves.
Like Artificial Ant, Lawnmower operates via side-effects and so execution order is important.
11.3 Where to Go Next

This is a woefully inadequate collection of resources that I’ve personally found useful.
11.3.1 Bibliographies, Surveys, and Websites

It’s an open secret that computer science researchers put a great many of their papers online, where
they’re often accessible from CiteSeer or CiteSeerx (CiteSeerx has newer documents, but the older
CiteSeer has a far superior interface. Go figure). Google Scholar is also useful, but usually points to
documents behind publisher’s firewalls.
http://citeseer.ist.psu.edu
http://citeseerx.ist.psu.edu
http://scholar.google.com
The Hitchhiker’s Guide to Evolutionary Computation was the FAQ for the Usenet group comp.ai.genetic.
Encore was its companion FAQ. Both of these resources are fairly dated: for example their software
collections don’t include anything current. Still, there’s a lot there, especially older work.
http://www.bionik.tu-berlin.de/EC/FAQ/www/
http://www.bionik.tu-berlin.de/EC/
http://code.google.com/p/hhg2ec/
The single biggest bibliography in the field is the Genetic Programming Bibliography, by Bill
Langdon, Steven Gustafson, and John Koza. I cannot overstate how useful this huge, immaculately
maintained bibliography has been to me (much of my work has been in genetic programming).
http://www.cs.bham.ac.uk/∼wbl/biblio/
Bill Langdon also maintains an extensive collection of bibliographies of EC conferences, etc.

http://www.cs.bham.ac.uk/∼wbl/biblio/ec-bibs.html
Carlos Coello Coello maintains a very large collection of multiobjective optimization resources.
http://www.lania.mx/∼ccoello/EMOO/
Tim Kovacs maintains a fairly complete bibliography on Learning Classifier Systems.

http://www.cs.bris.ac.uk/∼kovacs/lcs/search.html
213
Jarmo Alander built a bibliography of practically all Genetic Algorithm publications up to 1993.
ftp://ftp.cs.bham.ac.uk/pub/Mirrors/ftp.de.uu.net/EC/refs/2500GArefs.ps.gz
Many other bibliographies can be found at the Collection of Computer Science Bibliographies. Look
under the Artificial Intelligence, Neural Networks, and Parallel Processing subtopics.
http://liinwww.ira.uka.de/bibliography/
Liviu Panait and I wrote a large survey of cooperative multiagent learning, which includes a lot of
stuff on coevolution and its relationships to other techniques (like multiagent Q-learning).
http://cs.gmu.edu/∼eclab/papers/panait05cooperative.pdf
Liviu Panait and Sean Luke, 2005, Cooperative multi-agent learning: The state of the art,
Autonomous Agents and Multi-Agent Systems, 11, 2005
A good Particle Swarm Optimization website, with lots of resources, is Particle Swarm Central.
http://www.particleswarm.info
Marco Dorigo maintains one of the best Ant Colony Optimization websites out there, including
pointers to software, publications, and venues.
http://www.aco-metaheuristic.org
Pablo Moscato has a big web page on Hybrid Evolutionary- and Hill-Climbing (“Memetic”)
Algorithms.
http://www.densis.fee.unicamp.br/∼moscato/memetic home.html
Paula Festa and Mauricio Rensede maintain an annotated bibliography of GRASP literature.
http://www.research.att.com/∼mgcr/grasp/gannbib/gannbib.html
Lee Spector has a website on the Push language and publications.

http://hampshire.edu/lspector/push.html
Michael O’Neill maintains a website on Grammatical Evolution resources.

http://www.grammatical-evolution.org
Rainer Storn also maintains a website on Differential Evolution.

http://www.icsi.berkeley.edu/∼storn/code.html
Various papers on Guided Local Search may be found at Edward Tsang’s laboratory website:
http://www.bracil.net/CSP/gls-papers.html
214
ACM SIGEvo is the ACM’s special interest group on evolutionary computation. In addition to
sponsoring various major conferences and journals, they also have a newsletter, SIGEvolution.
The IEEE Computational Intelligence Society’s Evolutionary Computation Technical Committee
(IEEE-CIS-ECTC, phew) is the approximate equivalent for the IEEE.
http://www.sigevo.org
http://www.sigevolution.org
http://www.ieee-cis.org/technical/ectc/
Mark Wineberg and Steffen Christensen regularly do a lecture on statistics specifically for meta-
heuristics researchers. Mark keeps a PDF of the lecture slides on his home page.
http://www.cis.uoguelph.ca/∼wineberg/publications/ECStat2004.pdf
http://www.cis.uoguelph.ca/∼wineberg/
11.3.2 Publications
Ready for lots more? Thomas Weise’s 800-page, free open text Global Optimization Algorithms: Theory
and Application goes in-depth in a number of the topics covered here. It’s got a lot of formalism,
with analysis and descriptive applications, and well over 2000 references. Did I mention it’s free?
http://www.it-weise.de
As far as books go, I think the single best guide to the craft of stochastic optimization is How to Solve
It: Modern Heuristics,194 by Zbigniew Michalewicz and David Fogel. Fun to read, filled with stories
and examples, and covering a very broad collection of issues and topics.
Zbigniew Michalewicz and David Fogel, 2004, How to Solve It: Modern Heuristics, Springer
The best book on Ant Colony Optimization is Marco Dorigo and Thomas Stützle’s Ant Colony
Optimization.
Marco Dorigo and Thomas Stützle, 2004, Ant Colony Optimization, MIT Press
If you are interested in genetic programming, check out Genetic Programming: an Introduction by
Wolfgang Banzhaf, Peter Nordin, Robert Keller, and Frank Francone. It’s aging but still good.
Wolfgang Banzhaf, Peter Nordin, Robert E. Keller, and Frank D. Francone, 1998, Genetic
Programming: An Introduction, Morgan Kaufmann
A much newer Genetic Programming work is A Field Guide to Genetic Programming by Riccardo
Poli, Bill Langdon, and Nick McPhee, which has the added benefit of being free online if you’re too
cheap to buy the print copy! (Buy the print copy.)
Riccardo Poli, William B. Langdon, and Nicholas Freitag McPhee, 2008, A Field Guide to
Genetic Programming, Available in print from lulu.com
http://www.gp-field-guide.org.uk/
194 Thisbook’s name is adapted from a very famous book which revolutionized the use of algorithmic methods for
solving complex problems: George Pólya, 1945, How to Solve It, Princeton University Press.
215
Kalyanmoy Deb’s Multi-Objective Optimization Using Evolutionary Algorithms is a good text for
multiobjective optimization, but it’s expensive.
Kalyanmoy Deb, 2001, Multi-Objective Optimization using Evolutionary Algorithms, Wiley
Kenneth Price, Rainer Storn, and Jouni Lampinen’s Differential Evolution is likewise good but
expensive.
Kenneth Price, Rainer Storn, and Journi Lampinen, 2005, Differential Evolution: A Practical
Approach to Global Optimization, Springer
James Kennedy, Russell Eberhart, and Yuhui Shi’s seminal book on Particle Swarm Optimization is
Swarm Intelligence. Unfortunately this was a very poor choice of name: there was already a Swarm
Intelligence, published two years earlier, largely about Ant Colony Optimization. That one was by
Eric Bonabeau, Marco Dorigo, and Guy Theraulaz.195
James Kennedy, Russell Eberhart, and Yuhui Shi, 2001, Swarm Intelligence, Morgan Kaufmann
Eric Bonabeau, Marco Dorigo, and Guy Theraulaz, 1999, Swarm Intelligence: From Natural to
Artificial Systems, Oxford University Press
Though it is getting somewhat long in the tooth, Melanie Mitchell’s An Introduction to Genetic
Algorithms is still quite a good, well, introduction to genetic algorithms.
Melanie Mitchell, 1996, An Introduction to Genetic Algorithms, MIT Press
David Fogel’s Blondie24 recounts the development of a one-population competitive coevolutionary

algorithm to learn how to play checkers very strongly, and casts it in the context of artificial
intelligence in general.
David Fogel, 2001, Blondie24: Playing at the Edge of AI, Morgan Kauffman
Last, but far from least, Ken De Jong’s Evolutionary Computation: A Unified Approach puts not only
most of the population methods but a significant chunk of all of metaheuristics under one unifying
framework. It covers a lot of what we don’t cover here: the theory and analysis behind these topics.
Kenneth De Jong, 2006, Evolutionary Computation: A Unified Approach, MIT Press
11.3.3 Tools
There’s lots of stuff out there. Here’s just a few:
So let’s get the obvious one out of the way first. ECJ196 is a popular population-based
toolkit with facilities for parallel optimization, multiobjective optimization, and most representa-
tions, including genetic programming. ECJ is designed for large projects and so it has a somewhat
steep learning curve. But its author is very responsive, and unusually handsome as well. If
you meet this person in the street, you should give him a big hug. ECJ also dovetails with a
195 Believe
it or not, there’s now a third book which has foolishly been titled Swarm Intelligence!
196 ECJ
actually doesn’t stand for anything. Trust me on this. Though people have made up things like “Evolutionary
Computation in Java” or whatnot.
216
multiagent simulation toolkit called MASON. Both are in Java. ECJ’s web page points to a lot of
other Java-based systems, if ECJ’s too heavyweight for you.
http://cs.gmu.edu/∼eclab/projects/ecj/
http://cs.gmu.edu/∼eclab/projects/mason/
If you prefer C++, here are two particularly good systems. EO is an evolutionary computation
toolkit, and an extension, ParadisEO, adds single-state, parallel, and multiobjective optimization
facilities. A competitor, Open BEAGLE, also provides good evolutionary and parallel tools.
http://eodev.sourceforge.net/
http://paradiseo.gforge.inria.fr
http://beagle.gel.ulaval.ca
The Particle Swarm Optimization folks have coalesced around a single C file as a kind of reference
standard. It’s well written and documented.
http://www.particleswarm.info/standard pso 2007.c
11.3.4 Conferences
The big kahuna is the Genetic and Evolutionary Computation Conference, or GECCO, run by ACM
SIGEvo (http://www.sigevo.org). GECCO is the merging of the former GP and ICGA conferences.
It’s usually held in the United States, and has lots of smaller workshops attached to it.
If you’re an undergraduate student, I highly recommend that you submit to the GECCO
Undergraduate Student Workshop. It’s a great venue to show off your stuff, and they’re friendly and
encouraging. If you’re a graduate student and would like some tough feedback on your proposed
thesis work, a great pick is the GECCO Graduate Student Workshop, where you present your work in
front of a panel of luminaries who then critique it (and they’re not nice!). This is a good thing: better
to hear it in a friendly workshop than when you’re doing your proposal or thesis defense! Both
workshops are specially protected from the rest of the conference and run by people who really
care about you as a student.
The primary European conference is the International Conference on Parallel Problem Solving from
Nature, or PPSN. It’s not historically been very large but of unusually high quality.
The third major conference is the IEEE Congress on Evolutionary Computation, or CEC, held in various
spots around the world. It’s often quite large.
The three conferences above are dominated by evolutionary computation techniques. An alternative
conference for other methods is the Metaheuristics International Conference or MIC.
The oldest theory workshop, and almost certainly the most respected venue in the field,197 is the
venerable Foundations of Genetic Algorithms workshop, or FOGA, run by ACM SIGEvo, and usually
in the United States. It’s not just about the Genetic Algorithm any more, but rather about all kinds
of metaheuristics theory: indeed, in 2009 there wasn’t a single Genetic Algorithm paper in the whole
197 For example, I have twice chosen to publish at FOGA rather than in even our best journals. That’s not atypical.
217
workshop! FOGA is held every other year. The year that FOGA’s not held, an alternative theory
workshop has lately been hosted at Schloss Dagstuhl (http://www.dagstuhl.de) in Germany.
Europe is also host to the European Conference on Genetic Programming, or EuroGP, an alternative
conference focused, not surprisingly, on genetic programming.
Not to be outdone, the Genetic Programming Theory and Practice Workshop, or GPTP, is held each
year at the University of Michigan, and is a bit more theory-oriented than EuroGP.
Ant Colony Optimization also has its own conference apart from the big ones above: the International
Conference on Ant Colony Optimization and Swarm Intelligence or ANTS.198
Particle Swarm Optimization and Ant Colony Optimization folks, among others, have also lately
been attending the IEEE Swarm Intelligence Symposium or SIS.
The area of Evolvable Hardware (EH)199 concerns itself with the optimization of hardware designs:
circuits, antennas, and the like. This field often has a prominent showing at the NASA/ESA
Conference on Adaptive Hardware and Systems.
I would be remiss in not mentioning conferences in Artificial Life (ALife),200 the simulation of
biological processes. ALife has long been strongly associated with metaheuristics, and particularly
with evolutionary computation.201 Major ALife conferences include the International Conference on
the Simulation and Synthesis of Living Systems (or ALife), the European Conference on Artificial Life (or
ECAL), and From Animals to Animats: the International Conference on Simulation of Adaptive Behavior
(or SAB). ALife and ECAL are run by the International Society of Artificial Life (http://alife.org).
SAB is run by the International Society for Adaptive Behavior (http://www.isab.org.uk).
11.3.5 Journals
At this point, I think the three primary journals in the field are all evolutionary computation
journals: but they accept papers on all topics in metaheuristics (and indeed many of the seminal
non-EC metaheuristics papers are in these journals).
198 Annoyingly, this is not an acronym.
199 Evolvable Hardware is notable in that the fitness function is often done in real hardware. Here’s a famous story.
Adrian Thompson was an early Evolvable Hardware researcher who worked on optimizing computer circuits using
evolutionary algorithms. Adrian had access to early releases of the Xilinx XC6216 FPGA, a chip which was capable of
forming arbitrary circuits on-chip through the deft use of a grid of programmable gates. The evolutionary algorithm
performed fitness assessment by actually programming the chip with the given circuit, then testing its performance
on an oscilloscope. Problem is, when Adrian received the final optimized circuits, they were sometimes consisted of
disconnected circuits with various vestigial sections that didn’t do anything. But when he deleted these regions, the
circuit stopped working on the chip! It turns out that the early Xilinx chips given to Adrian had bugs on them, and
the evolutionary algorithm was finding solutions which identified and took advantage of the bugs. Not generalizable! See
Adrian’s homepage for various literature: http://www.informatics.sussex.ac.uk/users/adrianth/ade.html
200 ALife lies at the intersection of computer scientists interested in stealing ideas from biology, and biologists interested
in using computers for modeling. Since you’re probably in the former camp, allow me to suggest a recent text which
romps all over the area, everything from evolutionary neural networks to swarms to Lindenmayer systems: Dario
Floreano and Claudio Mattiuissi, 2008, Bio-Inspired Artificial Intelligence: Theories, Methods, and Technologies, MIT Press.
201 ALife is so strongly associated with evolutionary computation that the journal Evolutionary Computation has a sister
journal, Artificial Life, which is among the primary journals in ALife.
218
The oldest and (I think) the most respected journal in the field is Evolutionary Computation (MIT
Press), often nicknamed ECJ.202 Originally founded by Ken De Jong, Evolutionary Computation has a
long track record of strong theoretical publication and good empirical work.203 As artificial life and
metaheuristics have long been closely associated, Evolutionary Computation has a sister journal, also
by MIT press: Artificial Life.
IEEE Transactions on Evolutionary Computation (IEEE TransEC) is a first-rate, highly ranked journal
which has a bit more of an application and technical emphasis. My first solo journal publication was
in IEEE TransEC and it was a most pleasant publication experience. Because it’s an IEEE journal,
IEEE TransEC also benefits from a high Impact Factor, which isn’t something to be dismissed!
Genetic Programming and Evolvable Machines (GPEM) is a newer journal which emphasizes genetic
programming and evolvable hardware, but takes a wide range of papers. It’s well regarded and is
published by Springer.204 The GPEM editor also maintains a blog, listed below.
http://gpemjournal.blogspot.com/
11.3.6 Email Lists

There are plenty of email lists, but let me single out three in particular.
EC-Digest is a long-running mailing list for announcements of interest to the metaheuristics

community. It’s moderated and low-bandwidth.
http://ec-digest.research.ucf.edu/
The Genetic Programming Mailing List is an active discussion list covering GP.
http://tech.groups.yahoo.com/group/genetic programming/
The Ant Colony Optimization Mailing List is a relatively light discussion list mostly for announce-
ments regarding ACO.
https://iridia.ulb.ac.be/cgi-bin/mailman/listinfo/aco-list
http://iridia.ulb.ac.be/∼mdorigo/ACO/mailing-list.html
11.4 Example Course Syllabi for the Text

Weeks are numbered, and each week is assumed to be approximately four hours of lecture time.
Topics are organized in approximate order of significance and dependency. Note that the Combina-
torial Optimization, Coevolution, and Model Fitting sections make fleeting, nonessential reference
to one another. Rough chapter dependencies are shown in the Table of Contents (page 1).
202 I’m not sure if Ken De Jong has yet forgiven me giving my software the same acronym — I just didn’t know that
Evolutionary Computation sometimes had a Journal after it!
203 Truth in advertising: I’m presently on the Evolutionary Computation editorial board.
204 More truth in advertising: I’m on the editorial board of Genetic Programming and Evolvable Machines.
219
Simple Syllabus A lightweight one-semester course covering common algorithms and topics.
1. Introduction, Gradient-based Optimization Optional:

(Sections 0, 1) 5. Multiobjective Optimization (Section 7.1–7.3)
2. Single-State Methods (Sections 2.1–2.4) 6. Combinatorial Optimization (Sections 8.1–8.3.1)
3. Population Methods (Sections 3.1–3.2, 3.6) 7. Parallel Methods (Sections 5.1–5.3)
4. Representation (Sections 4.1, 4.3.1–4.3.3) 8. Coevolution (Sections 6.1 – 6.3)
Firehose Syllabus An intensive one-semester senior-level or masters’ level course.
1. Introduction, Gradient-based Optimization 7. Combinatorial Optimization (Section 8)

(Sections 0, 1) 8. Parallel Methods (Section 5)
2. Single-State Methods (Sections 2.1–2.6) 9. Coevolution (Section 6)
3. Population Methods (Sections 3.1–3.6) 10. Model Fitting (Section 9)
4. Representation (Sections 4.1, 4.2, and 4.5) 11. Policy Optimization (Sections 10.1–10.2)
5. Representation (Sections 4.3, 4.4, and 4.6) (presuming no prior knowledge of Q-Learning)
6. Multiobjective Optimization (Section 7) 12. Policy Optimization (Sections 10.3–10.5)
220
Errata
The errata205 omits a great many minor typo fixes and other insignificant changes.
Errata for Online Version 0.1 → Online Version 0.2

Page 0 Fixed the URL: it was “books” rather than “book”. It’s a long story.
Page 0 Updated the Thanks. I wasn’t sufficiently thankful.
Page 12 Added a thing on Newton.
Page 21 Added a thing on Gauss.
Page 22 Tweaked Table 1 in the hopes that it’s a bit less obtuse now.
Page 33 Added mention of adaptive and self-adaptive operators to Evolution Strategies.
Page 54 Algorithm 39 (Particle Swarm Optimization), line 3, should read:
β ← proportion of personal best to be retained

Page 79 New representation added (stack languages and Push). I’d meant to include them but hadn’t yet figured out
how to do it (they’re both trees and lists).
Page 214 Added Push to the Miscellany.
Page 221 Added the Errata. Let no one say I’m afraid of self-reference.
Thanks to Lee Spector, “Markus”, Don Miner, Brian Ross, Mike Fadock, Ken Oksanen, Asger Ottar Alstrup, and James
O’Beirne.

Page 0 Updated the Thanks. Again, not nearly thankful enough.
Page 13 Algorithm 3 (Newton’s Method with Restarts (One Dimensional Version)), line 6, should read:
until f ) ( x ) = 0 and f )) ( x ) < 0

Page 13 Added a footnote on zero gradient.
Page 117 Added a footnote on Alternating Optimization and its relationship to N-Population Cooperative Coevolution.
Page 213 Revised the URLs for the Hitchhiker’s Guide to Evolutionary Computation (the Usenet FAQ), and for Encore.
Thanks to Joerg Heitkoetter, Don Sofge, and Akhil Shashidhar.

Page 30 New footnote distinguishing between survival and parent selection.
Page 132 Algorithm 100 (Computing a Pareto Non-Dominated Front), after line 8, insert:
break out of inner for-loop

Page 134 Algorithm 102 (Multiobjective Sparsity Assignment), added a comment to make it clear how to use the
algorithm on a single Pareto Front Rank.
Page 134 Algorithm 102 (Multiobjective Sparsity Assignment), line 12 should read:
return R with Sparsities assigned

Page 135 Algorithm 104 (An Abstract Version of the Non-Dominated Sorting Genetic Algorithm II (NSGA-II)), added
some comments to make it more clear what AssessFitness does.
Page 213 Further updates to Hitchhiker’s Guide to Evolutionary Computation URL.
Page 220 Added PSO to the Simple Syllabus.
Thanks to Jeff Bassett, Guillermo Calderón-Meza, and Joerg Heitkoetter.
205 Man, you are really bored, aren’t you. Reading the errata. I mean, come on.
221
Page 24 Scatter Search should have been Section 3.3.5.
Page 36 Added footnote to note Crossover’s use with ES.
Page 39 Added a footnote to mention Schwefel’s early work in K-vector uniform crossover.
Page 40 Modified Footnote 26 to give more credit to Schwefel and discuss terminology.
Page 47 Complete revision of section to broaden definition of memetic algorithms.
Page 100 Added a bit more on Asynchronous Evolution.
Page 142 Added Footnote 120 on further reading on constrained stochastic optimization.
Page 155 Changed the Section name from Model Fitting to Optimization by Model Fitting. It’s more fitting.
Thanks to Hans-Paul Schwefel, Pablo Moscato, Mark Coletti, and Paul Wiegand.

Page 10 Added notation on functions.
Page 38 Added some further discussion about the perils of crossover.
Page 43 Added a footnote on tournament selection variations.
Page 45 Minor extension in discussion about elitism.
Page 47 More discussion about memetic algorithms.
Page 73 Split the footnote.
Pages 114–119 Changed “K-fold” to “K-fold” in all algorithm names.
Page 115 Because it’s not formally elitist, changed the name of Algorithm 87 (Elitist Relative Fitness Assessment
with an Alternative Population) to K-fold Relative Fitness Assessment with the Fittest of an Alternative
Population.
Page 119 Changed “k more tests” to simply “k tests” in the comments (Algorithm 91, K-fold Joint Fitness Assessment
of N Populations).
Page 199 Changed recommendation to use 100 rather than ∞ for large degrees of freedom.
Page 215 Added Global Optimization Algorithms: Theory and Application. How did I manage to not include this?
Thanks to Yury Tsoy and Vittorio Ziparo.

Page 25 Improved the code of Algorithm 15 (Feature-based Tabu Search) (no corrections).
Page 74 Improved the code of Algorithm 56 (The PTC2 Algorithm) (no corrections).
Page 96 Algorithm 68 (Simple Parallel Genetic Algorithm-style Breeding), line 2, should read:
T ← set of threads { T1 , ...Tn }

Page 172 Text should read:
We need to replace the ∑s) P(s) |s, a) portion of Equation 1.

Page 176 Text should read:
$4, 1%
Page 175 Equation should read
∞
Q∗ (s, a) = R(s, a) + γ ∑ P(s) |s, a) max E[ ∑ γt R(st , at )|s0 = s) , a0 = a) , at≥1 = π ∗ (st )]
s) a) t =0
Page 178 Added another reference to K-Nearest Neighbor.
222
Page 191 Algorithm 131 (XCS Fitness Updating), line 21, should read:
for each rule Ai ∈ A do

Page 192 Line number references (“28 and 29”) corrected.
Page 193 Algorithm 133 (XCS Fitness Updating (Extended)), line 15, should read:
Experience(Ai ) ← Experience(Ai ) +1
Page 193 Algorithm 133 (XCS Fitness Updating (Extended)), line 23, should read:
for each rule Ai ∈ A do

Page 193 Line number references (“15, 16, and 24”) corrected.
Thanks to Yury Tsoy.

Page 25 Algorithm 15 (Feature-based Tabu Search), line 9, should read:
Remove from L all tuples of the form $ X, d% where c − d > l

Page 85 Associated agents with policies.
Page 89 Algorithm 62 (Simple Production Ruleset Generation), completely updated to fix a few minor bugs and to
give the user the option to prevent disconnected rules. The revised algorithm now reads:
1: !t ← pre-defined set of terminal symbols (that don’t expand)

2: p ← approximate probability of picking a terminal symbol
3: r ← flag: true if we want to allow recursion, else false
4: d ← flag: true if we want to allow disconnected rules, else false
5: n ← a random integer > 0 chosen from some distribution

6: !v ← vector of unique symbols $v1 , ..., vn % " The symbol in v1 will be our “start” symbol
−−→
7: rules ← empty vector of rules $rules1 , ..., rulesn %
8: for i from 1 to n do " Build rules
9: l ← a random integer ≥ 1 chosen from some distribution
10: !h ← empty vector of symbols $h1 , ...hl %
11: for j from 1 to l do
12: if (r = false and i = n) or p < random value chosen uniformly from 0.0 to 1.0 inclusive then
13: h j ← a randomly chosen terminal symbol from !t not yet appearing in !h
14: else if r = false then
15: h j ← a randomly chosen nonterminal from vi+1 , ..., vn not yet appearing in !h
16: else
17: h j ← a randomly chosen nonterminal symbol from !v not yet appearing in !h
18: rulesi ← rule of the form vi → h1 ∧ h2 ∧ ... ∧ hl
19: if d = false then " Fix disconnected rules
21: if vi does not appear in the head of any of the rules rules1 , ..., rulesi−1 then
22: l ← a random integer chosen uniformly from 1 to i − 1 inclusive
23: Change rulesl from the form vl → h1 ∧ ... to the form vl → vi ∧ h1 ∧ ...
−−→
24: return rules
Page 120 Michael Jordan was a guard, not a center. That was embarrassing!
Page 134 Algorithm 102 (Multiobjective Sparsity Assignment) tweaked to permit objectives with different ranges — this
was added by Deb et al in later implementations of the algorithm. To do this, insert after line 2 the following
line:
Range(Oi ) function providing the range (max − min) of possible values for a given objective Oi
223
Then change line 12 to read:
ObjectiveValue(Oi , Fj)+1 ) − ObjectiveValue(Oi , Fj)−1 )

Sparsity(Fj) ) ← Sparsity(Fj) ) +
Range(Oi )
Page 167 Again associated agents with policies.
Page 197 Moved Footnote 171.
Page 215 Moved Mark and Steffan’s resource entry to the end of the list to be consistent with Footnote 176.
Page 218 Added Footnote 199 about Adrian Thompson and computer circuits.
Page 218 Added Footnote 200 about Dario’s book.
Thanks to Ivan Krasilnikov, Yuri Tsoy, Uday Kamath, Faisal Abidi, and Yow Tzu Lim.

All Pages Added hyperlinks to all chapter, page, remote footnote, or index references. They appear as little red
rectangles. Footnote marks do not have them: it’s too distracting. The hyperrefs do not appear when the
publication is printed. We’ll see if people can stand them or not.
Page 0 Noted Roger Alsing and added him to the index. He wasn’t there due to a bug in LATEX: you can’t add index
entries for pages less than 0.
Page 7 Footnote 1 expanded to discuss more alternate names.
Page 9 Reformatted the Notation to be entirely in itemized form.
Page 19 Line 14 of Algorithm 10 (Hill Climbing with Random Restarts) should read:
Until Best is the ideal solution or we have run out of total time
Page 22 It’s the Box-Muller-Marsaglia Polar Method.
Page 22 Fixes to Algorithm 12 (Sample from the Gaussian Distribution (Box-Muller-Marsaglia Polar Method)). We
were doing normal distribution transforms using variance instead of standard deviation. Specifically, Lines 8
and 9 should read:
'
g ← µ + xσ −2 lnww
and
'
h ← µ + yσ −2 lnww
Page 22 We were doing normal distribution transforms using variance instead of standard deviation. Equation should
read should read:
√
N (µ, σ2 ) = µ + σ2 N (0, 1) = µ + σN (0, 1)
Page 23 Line 11 of Algorithm 12 (Simulated Annealing) should read:
until Best is the ideal solution, we have run out of time, or t ≤ 0

Page 27 Line 16 of Algorithm 16 (Iterated Local Search with Random Restarts) should read:
Until Best is the ideal solution or we have run out of total time
Page 35 Line 16 of Algorithm 20 (The Genetic Algorithm) should read:
Q ← Q ∪ {Mutate(Ca ), Mutate(Cb )}
Page 38 Footnote 24 added to discuss epistasis.
Page 38 Expansion of the paragraph discussing crossover and linkage.
Page 50 Line 2 of Algorithm 37 (Simplified Scatter Search with Path Relinking) is deleted (it defined an unused
variable).
224
Page 83 Line 7 of Algorithm 9 (One Point List Crossover) should read:
!y ← snip out wd through wk from w

!
Thanks to Daniel Carrera, Murilo Pontes, Maximilian Ernestus, Ian Barfield, Forrest Stonedahl, and Yuri Tsoy.

Page 61 Line 12 of Algorithm 42 (Random Walk Mutation) should read:
until b < random number chosen uniformly from 0.0 to 1.0 inclusive
Page 66 Algorithm 47 renamed to Build a Simple Graph.
Page 67 Algorithm 48 completely replaced and renamed to Build a Simple Directed Acyclic Graph. The revised algorithm
now reads:
2: D (m) ← probability distribution of the number of edges out of a node, given number of in-nodes m

9: p ← random integer ≥ 1 chosen using D (i − 1)
10: for j from 1 to p do
11: repeat
12: k ← random number chosen uniformly from 1 to i − 1 inclusive
13: until f (i, k, N, E) returns ’true’
14: g ← new edge from Ni to Nk
15: ProcessEdge(g)
16: E ← E ∪ { g}
17: return N, E
Page 73 Line 11 of Algorithm 54 (The Full Algorithm) should read:
Child i of n ← DoFull(depth + 1, max, FunctionSet)

Page 74 Line 9 of Algorithm 56 (The PTC2 Algorithm) should read:
for each child argument slot b of r

Page 82 Line 10 of Algorithm 58 (Random Walk) should read:
until b < random number chosen uniformly from 0.0 to 1.0 inclusive
Page 94 Line 37 of Algorithm 65 (Thread Pool Functions) should read:
Wait on l
Page 101 Line 21 of Algorithm 73 (Asynchronous Evolution) should read:
if || P|| = popsize
Page 110 Line 7 of Algorithm 81 (Single-Elimination Tournament Relative Fitness Assessment) should read:
Q j defeated Q j+1 in that last Test

Page 116 Text changed to clarify that Potter and DeJong proposed cooperative coevolution, not competitive coevolu-
tion.
Page 218 Footnote 201 added about the relationship between ALife and Evolutionary Computation.
Thanks to Gabriel Balan and Muhammad Iqbal.
225
Page 51 Added some new discussion on survival selection.
Page 52 Algorithm 38 (Differential Evolution) completely revised to fix a bug and to make the resulting algorithm
simpler. The new version is:
1: α ← mutation rate " Commonly between 0.5 and 1.0, higher is more explorative
3: P←$% " Empty population (it’s convenient here to treat it as a vector), of length popsize
4: Q←! " The parents. Each parent Qi was responsible for creating the child Pi
6: Pi ← new random individual
7: Best ← !
8: repeat
11: if Q 4= ! and Fitness(Qi ) > Fitness(Pi ) then
12: Pi ← Qi " Retain the parent, throw away the kid
14: Best ← Pi
15: Q←P
16: for each individual Qi ∈ Q do " We treat individuals as vectors below
17: !a ← a copy of an individual other than Qi , chosen at random with replacement from Q
18: !b ← a copy of an individual other than Qi or !a, chosen at random with replacement from Q
19: !c ← a copy of an individual other than Qi , !a, or !b, chosen at random with replacement from Q
20: d! ← !a + α(!b −!c) " Mutation is just vector arithmetic
21: Pi ← one child from Crossover(d,! Copy(Qi ))
22: until Best is the ideal solution or we ran out of time
23: return Best
Page 102 Algorithm 75 renamed to Random Walk Selection. Text immediately before the algorithm changed to properly
reflect the description of the algorithm. Finally, Line 11 of the algorithm should read:
return the individual located at !l in the space

Page 123 Comment on Line 6 of Algorithm 92 (Implicit Fitness Sharing) should read:
Ri,j is individual Ri ’s sum total reward for Tj

Page 123 Lines 10 and 11 of Algorithm 92 (Implicit Fitness Sharing) should read:
for each individual Ql ∈ Q do

i ← index of Ql in P)
Page 129 Line 6 of Algorithm 94 (Multiobjective Lexicographic Tournament Selection) should read:
for j from 1 to n do
Page 139 For clarity, added a line and revised a comment in Algorithm 107 (An Abstract Version of the Strength Pareto
Evolutionary Algorithm 2 (SPEA2)).
Page 174 The wrong equation had been labelled Equation 3.
Page 170 Reference to Figure 64 changed to Figure 63.
Page 182 Line 41 of Algorithm 124 (S AMUEL Fitness Assessment) should read:
if dofitness is true then

Page 190 Comment on Line 2 of Algorithm 130 (XCS Action Selection) should read:
0≤%≤1
226
Page 194 Caption to Figure 66 should read:
A robot world with three rooms, a door, and a switch. available actions for each room are shown. The robot
can only exit if the door is opened. Flicking the switch opens the door.
Page 212 Changed the Lawnmower example to more clearly indicate that (frog i) is not in ADF2; and that the ADF2
and ADF1 are reordered and renamed with respect to Koza’s originals.
Thanks to Joseph Zelibor and Muhammad Iqbal.

All Pages Made the term Evolution Strategies (and ES) plural. I always view it as a mass noun, and thus singular, but
I’m in the minority there.
Page 39 Additional line inserted at Line 11 of Algorithm 27 (Uniform Crossover among K Vectors), which reads:
w
! ← Wj
Page 41 Explained what it means to “select with replacement”.
Page 51 It’s Jouni Lampinen.
Page 60 Added summary of vector representation functions discussed so far.
Page 71 Expanded the C and Lisp code, removing the value of a footnote.
Page 93 Added a reference to Zbigniew Skolicki’s thesis.
Page 208 The problems ZDT1, ZDT2, and ZDT3 should have the range:
xi ∈ [0, 1]
ZDT4 should have the range:
x1 ∈ [0, 1], xi>1 ∈ [−5, 5]

Thanks to Petr Pos̆ı́k and Faisal Abidi.

Only minor modifications. Buy the Book! http://cs.gmu.edu/∼sean/book/metaheuristics/
Thanks to Kevin Molloy.
Errata for Online Version 1.0 (First Print Edition) → Online Version 1.1
(Note: many of these errata found their way into later versions of the first Print Edition, after January 1, 2011)
Page 44 Added small item about handling odd population sizes with Elitism.
Page 60 Added page numbers to algorithm references in table.
Page 82 Added footnote on better handling of Grammatical Evolution.
Page 206 Modified traditional bounds for Rosenbrock to xi ∈ [−2.048, 2.048]. Also adjusted the figures in Figure 67 to
make Rosenbrock more easily understood with the revised bounds.
Page 206 Rastrigin’s function should read:
f ($ x1 , ..., xn %) = 10n + ∑in=1 xi2 − 10 cos(2πxi )

Additionally, Rastrigin wasn’t part of the DeJong Test Suite: though it often has traditional bounds of
xi ∈ [−5.12, 5.12] in the literature.
Page 206 Modified traditional bounds for Schwefel to xi ∈ [−512.03, 511.97].
Page 216 Revised Journi Lampinen to Jouni Lampinen (again).
Thanks to Keith Sullivan, Matthew Molineaux, and Brian Olson.
227
228
Index
%-greedy action selection, 173, 190 bin packing, 141
(µ+λ), 32 black box optimization, 7
(µ+1), 46 bloat, 83, 90
(µ, λ), 31 Blondie24, 106
(1+λ), 22 Bonabeau, Eric, 216
(1+1), 21 Bonferroni correction, 202
(1, λ), 22 bootstrapping, 172
Born, Joachim, 206
Ackley, David, 37, 203, 204 Box, George Edward Pelham, 22
action set, 186 Box-Muller-Marsaglia Polar Method, 22
actions, 168 breeding, 29
activity level, 181 Brindle, Anne, 43
Agarwal, Sameer, 135 building blocks, 38, 210
agent, 85, 167, 168 Butz, Martin, 189
Alander, Jarmo, 214
aliased states, 194 candidate solution, see individual, 15
allele, 29 Cantú-Paz, Eric, 166
Alsing, Roger, 0 Caruana, Rich, 161
Alternating Optimization (AO), 117 Cellular Encoding, 78
Andre, David, 78 Chellapilla, Kumar, 105, 106
Angeline, Peter, 65 child, 29
annealing, 24 Chinook, 107
ANOVA, 202 Christensen, Steffen, 201, 215
Ant Colony Optimization (ACO), 146 chromosome, 29
Ant Colony System (ACS), 150 classification, 155, 176
Ant System (AS), 147 closure, see operator, closed
ant trails, 146 co-adaptive, 104
AQ, 156 Coello Coello, Carlos, 213
arbitration scheme, 86, 176 coevolution, 103
archive, 120, 135 N-Population Cooperative, 103, 116
arity, 72 1-Population Competitive, 103, 105
arms race, 116 2-Population Competitive, 103, 111
arrays, 10 parallel, 113
artificial immune systems, 123 parallel previous, 114
Artificial Life (ALife), 218 sequential, 112
Asada, Minoru, 169 serial, 112
assessment procedure, 15 compositional, 103
Asynchronous Evolution, 100 test-based, 103
automatically defined functions (ADFs), 76 collections, 9
automatically defined macros (ADMs), 77 Collins, J. J., 81
combinatorial optimization problem, 141
Baker, James, 42 Compact Genetic Algorithm (cGA), 162
Baldwin Effect, 49 compactness, 79
Baluja, Shumeet, 161 components, 141
Banzhaf, Wolfgang, 80, 215 computational effort, 197
Baxter, John, 26 cons cells, 80
Bayes Network, 165 convergence, 38
Bayesian Optimization Algorithm (BOA), 166 convergence time, 11
Bellman Equation, 171 Copy, 15, 57
Bellman, Richard, 171 copy-forward, 97
Bennett, Forrest, 78 cost, 142
best of run, 197 covariance matrix, 160
biasing, 30, 60 cover, 176
229
Cramer, Nichael, 70 Estimation of Distribution Algorithms (EDAs), 155, 158
credit assignment, 120 Multivariate, 165
crossover, 29, 31, 36 Univariate, 161
Clustered, 184 evaluation, 29
Intermediate Recombination, 40 evaporation, 147
for Integers, 62 Evolution Strategies (ES), 31
Line Recombination, 39 Evolutionary Algorithm (EA), 29
for Integers, 62 Evolutionary Computation (EC), 29
Multi-Point, 37 Evolutionary Programming (EP), 34
One-Point, 36 Evolvable Hardware (EH), 218
List, 83 Expectation Maximization (EM), 117, 165
Subtree, 74 explicit speciation, 121
Two-Point, 36 Exploration versus Exploitation, 18, 20, 173
List, 83 external state, 168, 194
Uniform, 36
Among K Vectors, 39 Feature-based Tabu Search, 25, 152
Crowding, 124 Feo, Thomas, 145
Deterministic, 124 Festa, Pauola, 214
cycles, 107 Fisher, Ronald Aylmer, 200
fitness, see quality, 29
Dawkins, Richard, 48 absolute, 104
De Jong, Kenneth, 46, 116, 118, 205, 216, 219 baseline, 184
Deb, Kalyanmoy, 133, 135, 198, 207, 216 external, 105
deceptive functions, 20, 57, 204 internal, 105
decision trees, 155 joint, 117
decoding, 58 relative, 104
demes, 97, 103, 111 fitness assessment, 29
desirability, 149 relative, 107
Differential Evolution (DE), 51 fitness functions, see problems
diploid, 112 fitness landscape, 29
directed acyclic graph, 67 fitness scaling, 43
directed mutation, 29, 53 fitness sharing, 122
Discipulus, 80 implicit, 123
distance measure, 122 Floreano, Dario, 218
distributions, 155 Fogel, David, 105, 106, 215
bivariate, 165 Fogel, Lawrence, 34, 105
Gaussian, 21 forest, 76
marginal, 155, 160 Forrest, Stephanie, 123
normal, 21 FORTH, 79
standard normal, 22 Francone, Frank, 80, 215
Diversity Maintenance, see niching full adjacency matrix, 64
Dorigo, Marco, 146, 150, 214–216 Full algorithm, 73
duplicability, 196 function set, 72
dynamic programming, 172 functions, 10
Eberhart, Russell, 53, 216 Gambardella, Luca, 150

Edge Encoding, 78 Gauss, Karl Friedrich, 21
Edwards, Howard, 197 Gelatt, Charles Daniel Jr., 23
elites, 44 gene, 29
elitism, 150 generalizability, 197, 218
encoding, 58 generation, 29
developmental, 64, 79 Generation Gap Algorithms, 46
direct, 64 generational, 29
indirect, 64, 79, 81, 86 generative models, 159
ephemeral random constant, 74 Genetic Algorithm (GA), 34
epistasis, 36, 38 Genetic Programming (GP), 70, 194
230
GENITOR, 45, 46 internal state, 194
genome, 29 introns, 90
genotype, 29, 57 invalid solutions, 141
Gibbs Sampling, 7 inviable code, 90
global optima, 12 island models, 97
global optimization algorithm, 13, 18 asynchronous, 98
Glover, Fred, 24, 49 synchronous, 98
GNARL, 65 island topology, 97
Goldberg, David, 122, 162, 166 fully-connected, 97
Gosset, William Sealy, 200 injection model, 97
Gradient Ascent, 11 toroidal grid, 97
Gradient Descent, 11 Iterated Local Search (ILS), 26, 142
Grammatical Evolution (GE), 81
graphs, 63 Jaśkowski, Wojciech, 110
Gray code, 59 Join, 30
Gray, Frank, 59 Jordan, Michael
Greedy Randomized Adaptive Search Procedures (GRASP), the basketball player, 120
145 the professor, 176
Grefenstette, John, 180, 184
Griewank, Andreas, 206 k-fold cross validation, 197
Grow algorithm, 72 k-means clustering, 117
Guided Genetic Algorithm, 154 k-Nearest-Neighbor (KNN), 155, 178
Guided Local Search (GLS), 152 Kauth, Joan, 45, 46
Gustafson, Steven, 213 kd-tree, 159
Keane, Martin, 78
Hamming cliff, 58 Keijzer, Martin, 80
Hamming distance, 122 Keller, Robert, 80
hard constraints, 143 Kennedy, James, 53, 216
Harik, Georges, 124, 162 kernelization, 179
Hessian, 12 Kirkpatrick, Scott, 23
heuristic, 142 Kitano, Hiroaki, 86
Hierarchical Bayesian Optimization Algorithm (hBOA), Klein, Jon, 80
166 Koch Curve, 87
Hill-Climbing, 8, 15 Kovaks, Timothy, 213
Hill-Climbing with Random Restarts, 19 Koza, John, 46, 70, 78, 209, 211, 213
Hillis, Daniel, 112 Krawiec, Krysztof, 110
history, 194
Holland, John, 34, 185 L-Systems, see Lindenmayer Systems
homologous, 38, 75 Laguna, Manuel, 49
Hornby, Gregory, 88 Lamarck, Jean-Baptiste, 48
hypervolume, 128, 198 Lamarckian Algorithms, 48
hypothesis, see model Lampinen, Jouni, 51, 216
hypothesis test, 198 Langdon, William, 210, 211, 213, 215
nonparametric, 200 Lanzi, Pier Luca, 189, 190
laziness, 120
iCCEA, 120 Learnable Evolution Model (LEM), 49, 155
illegal solutions, 141 learning bias, 156
incest prevention, 121 Learning Classifier Systems (LCS), 167, 177, 185
individual, see candidate solution, 29 learning gradient, 105
induction, 155 learning rate, 150
infeasible solutions, 141 Lindemayer Systems, 218
informant, 53 Lindenmayer Systems, 87
informative gradient, 20 Lindenmayer, Aristid, 87
initialization, 30, 57 Linear Genetic Programming, 80
adaptive, 185 linear problem, 203
initialization procedure, 15 linkage, 36, 38, 203
231
lists, 70, 80 Bit-Flip, 36
Lobo, Fernando, 162 Creep, 184
local optima, 12 Duplicate Removal, 85
local optimization algorithm, 12, 18 Gaussian Convolution, 21, 33
loss of gradient, 116 Gaussian Convolution Respecting Zeros, 65
Lourenço, Helena, 26 Integer Randomization, 61
Lucas, Simon, 91 Point, 62
Random Walk, 61
machine learning, 155 Subtree, 75
Mahfoud, Samir, 124 mutation rate, 33
Manhattan distance, 133 adaptive, 33
Manhattan Project, 23
Markov Chain Monte Carlo (MCMC), 7 NEAT, 65, 107
Markov Decision Processes (MDPs), 42 Needle in a Haystack style functions, 20, 57
Markov, Andrey Andreyevich, 169 neighbors, 102
Markovian environment, 169 NERO, 107
Marsaglia, George, 22 neural networks, 155
Martı́, Rafael, 49 Newton’s Method, 12
Martin, Olivier, 26 Newton’s Method with Restarts, 13
master-slave fitness assessment, 99 Newton, Sir Isaac, 12
match score, 177 Ng, Andrew, 176
match set, 85, 180, 186 niches, 104
matrices, 10, 63, 70, 78 niching, 104
Mattiussi, Claudio, 218 No Free Lunch Theorem (NFL), 203
maxima, 11 noisy functions, 19
McPhee, Nicholas, 215 Non-Dominated Sorting, 133
mean vector, 160 Non-Dominated Sorting Genetic Algorithm II (NSGA-II),
Memetic Algorithms, 48 135
memory, 194 non-homologous, see homologous
Messom, Chris, 197 Nordin, Peter, 80, 215
metaheuristics, 7 null hypothesis, 198
metric distance, 122
Metropolis Algorithm, 23 O’Neill, Michael, 81, 214
Metropolis, Nicholas, 23 objective, 127
Meyarivan, T., 135 objective functions, see problems
Michalewicz, Zbigniew, 142, 215 One-Fifth Rule, 34
Michalski, Ryszard, 155 operator
Michigan-Approach Learning Classifier Systems, 86, 167, adaptive, 33
177, 185 closed, 70, 81, 143
microclassifiers, 192 self-adaptive, 33, 80, 185
Miikkulainen, Risto, 65, 107 Opportunistic Evolution, 100
minima, 11 over-specification, 86, 176
miscoordination, 120
Mitchell, Melanie, 216 Pólya, George, 215
model, 10, 155 Panait, Liviu, 91, 120, 165, 197, 214
discriminative, 156 parent, 29
generative, 156 Pareto domination, 127
modification procedure, 15 Pareto front, 127
modularity, 76, 79, 86 concave, 127
Mona Lisa, 0 convex, 127
Montana, David, 77 discontinuous, 128
Monte Carlo Method, 23 local, 128
Moscato, Pablo, 48, 214 nonconvex, 127
Mühlenbein, Heinz, 39, 162, 206 Pareto Front Rank, 132
Muller, Mervin, 22 Pareto nondominated, 127
mutation, 29, 31 Pareto strength, 136
232
Pareto weakness, 136 ZDT1, 207
Pareto wimpiness, 136 ZDT2, 208
parse trees, 71 ZDT3, 208
parsimony pressure, 91 ZDT4, 208
double tournament, 92 Prusinkiewicz, Przemyslaw, 87
lexicographic, 91 PTC2, 73
linear, 91 Push, 80
non-parametric, 91
Particle Filters, 42 Q-learning, 167, 169, 172
Particle Swarm Optimization (PSO), 29, 53 Q-table, 169
particles, 53 Q-value, 170
Pelikan, Martin, 166 quadtree, 159
penalties, 153 quality, see fitness, 15
Perelson, Alan, 123 queues, 9
phenotype, 29, 57
pheromone, 146 Ramped Half-and-Half algorithm, 73
Pitt-Approach Rule Systems, 86, 167, 180 Ramsey, Connie, 180
Poli, Riccardo, 211, 215 random number generator, 195
policy, 85, 167, 168 java.util.Random, 195
policy search, 176 linear congruential, 195
Pollack, Jordan, 65 Mersenne Twister, 195
population, 29 RANDU, 195
alternative, 111 Random Search, 8, 19
collaborating, 111 random walk, 23
foil, 111 Rastrigin, Leonard Andreevich, 206
primary, 111 Rechenberg, Ingo, 31, 205
Population-Based Incremental Learning (PBIL), 161 recombination, see crossover
PostScript, 79 REINFORCE, 176
Potter, Mitchell, 116, 118 reinforcement, 167
Pratap, Amrit, 135 negative, 168
prediction, 189 positive, 168
prediction error, 189 reinforcement learning, 167, 172
premature convergence, 32, 38 multiagent, 120
Price, Kenneth, 51, 216 relative overgeneralization, 120
probability distributions, 10 Rensende, Mauricio, 214
problem, 15 replicability, 195
problems representation, 17, 57
11-bit Boolean Multiplexer, 209 resampling techniques, 29
Artificial Ant, 71, 211 Resende, Mauricio, 145
De Jong test suite, 204–206 reward, see reinforcement
Even N-Parity, 210 Richardson, Jon, 122
Griewank, 206 Robert Keller, 215
Knapsack, 141 Robinson, Alan, 80
Lawnmower, 212 robustness, 105
Leading Ones, 203 Rosenbluth, Arianna and Marshall, 23
Leading Ones Blocks, 203 Rosenbrock, Howard, 206
Linear Problems, 204 rule
Max Ones, 203 active, 181
OneMax, 203 default, 86, 177
Rastrigin, 206 production, 84
Schwefel, 206 state-action, 84, 85
Sphere, 205 rule body, 84, 177
Step, 204 rule covering, 183
Sum, 204 rule deletion, 183
Symbolic Regression, 71, 72, 74, 77, 209 rule generalization, 183
Traveling Salesman (TSP), 25, 141 rule head, 84, 177
233
rule merging, 183 states, 168
rule specialization, 183 statistically significant, 200
rule strength, 181 steady-state, 29, 45
Ryan, Conor, 81, 91 Steepest Ascent Hill-Climbing, 16
Steepest Ascent Hill-Climbing with Replacement, 16
saddle points, 11 Stewart, Potter, 7
sample, 29 stochastic, 16
sample distributions, 158 stochastic optimization, 7
sampling stochastic processes, 169
region-based, 157 stochastic search, 7
rejection, 156 Storn, Rainer, 51, 214, 216
weighted rejection, 157 Strength Pareto Evolutionary Algorithm 2 (SPEA2), 136
SAMUEL, 49, 180 strings, see lists
Saunders, Gregory, 65 Student’s t-Test, 200
Scatter Search with Path Relinking, 49 subpopulations, 97, 103
schedule, 23 subsolution, 117
schema theory, 38 subsumption, 193
Schlierkamp-Voosen, Dirk, 39 subtree selection, 76
Schoenauer, Marc, 142 supervised learning, 177
Schomisch, M., 206 Support Vector Machines (SVMs), 155, 179
Schultz, Alan, 180 swarm, 53
Schwefel, Hans-Paul, 31, 39, 40, 206 symbols, 88
seeding, 30, 60 nonterminal, 86
selection, 29 terminal, 86
Fitness-Proportionate, 41
Fitnessless, 110 tabu list, 24
non-parametric, 43 Tabu Search, 24, 142
parent, 30, 51 Teller, Augusta and Edward, 23
Roulette, 41 Teller, Eric (Astro), 23
Stochastic Universal Sampling, 42 temperature, 24
survival, 30, 51 test cases, 105, 111, 197
Tournament, 43 test problems, see problems
Restricted, 124 test set, 197
Truncation, 41 tests, 105
selection pressure, 22 Theraulaz, Guy, 216
selection procedure, 15 Thiele, Lothar, 198, 207
Shi, Yuhui, 216 Thompson, Adrian, 218
Sigvardsson, Oskar, 0 threads, 95
Simulated Annealing, 23 Tinsley, Marion, 107
Skolicki, Zbigniew, 93, 139 tournament size, 43
Smith, Robert, 123 training set, 197
smoothness, 20, 57 transition model, 168
Solkoll, 0 Tree-Style Genetic Programming, 46
sorting networks, 112 trees, 70
sparsity, 133 Truncation Selection, 31
spatially embedded models, 101 Tsang, Edward, 152, 214
species, 104 Tukey, John, 202
specificity, 176 tuples, 9
Spector, Lee, 77, 78, 80, 214 Tweak, 15, 18, 57
spread, 128 type constraints, 77
Srinvas, N., 133 typing
Stützle, Thomas, 26, 146, 215 atomic, 77
stack languages, 79, 81 polymorphic, 78
Stanley, Kenneth, 65, 107 set, 77
state space, 86
state-action rules, 167 unbiased estimator, 164
234
under-specification, 86, 177
unimodal functions, 19
Univariate Marginal Distribution Algorithm (UMDA), 162
utility, 168
utility error, 178, 189
utility variance, 177, 181
value, 142, 168

Vecchi, Mario, 23
vector processor, 102
vectors, 9
Vo, Christopher, 165
Voronoi tessellation, 178
vote, 86
Voudouris, Chris, 152
Walker, Matthew, 197

weak methods, 7
weights, 142
Whitley, Darrell, 45, 46
Wiegand, Paul, 120
Wieloch, Bartosz, 110
Williams, Ronald, 176
Wilson, Stewart, 186
Wineberg, Mark, 201, 215
XCS, 189
Xilinx, 218
Zeroth Level Classifier System (ZCS), 186

Ziggurat Method, 22
Zitzler, Eckart, 198, 207
235

Essentials of Metaheuristics by Sean Luke

Uploaded by

Copyright:

Available Formats

Essentials of Metaheuristics by Sean Luke

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Essentials of Metaheuristics by Sean Luke

Uploaded by

Copyright:

Available Formats

Essentials of Metaheuristics

A Set of Undergraduate Lecture Notes by

Figure 0 The Mona Lisa, esti-

Get the latest version of this document or suggest improvements here:

Cite this document as:

BibTEX: @Book{ Luke2009Metaheuristics,

2 Single-State Methods Depends on Section 1 15

3 Population Methods Depends on Section 2 29

4 Representation Depends on Sections 2 and 3 57

5 Parallel Methods Depends on Section 3 93

6 Coevolution Depends on Section 3 103

7 Multiobjective Optimization Depends on Section 3 127

8 Combinatorial Optimization Depends on Sections 2 and 3 141

10 Policy Optimization Depends on Sections 3 and 4 167

11 Miscellany Depends on Section 4 195

0.1 What is a Metaheuristic?

Algorithm 0 Bubble Sort

Algorithm 1 Gradient Ascent

Algorithm 2 Newton’s Method (Adapted for Maximum Finding)

Because it employs the second derivative, Newton’s

Algorithm 3 Newton’s Method with Restarts (One Dimensional Version)

2: S ← some initial candidate solution

2: S ← some initial candidate solution

Algorithm 7 Generate a Random Real-Valued Vector

3: !v ← a new vector $v1 , v2 , ...vl %

Algorithm 8 Bounded Uniform Convolution

Figure 6 Four example quality functions.

2.2 Single-State Global Optimization Algorithms

Algorithm 9 Random Search

Algorithm 10 Hill-Climbing with Random Restarts

2: S ← some initial random candidate solution

• Use a Large Sample Try many candidate solutions in parallel.

Algorithm 12 Sample from the Gaussian Distribution (Box-Muller-Marsaglia Polar Method)

7: until 0 < w '

2: S ← some initial candidate solution

2.5 Tabu Search

Algorithm 14 Tabu Search

3: S ← some initial candidate solution

3: S ← some initial candidate solution

2.6 Iterated Local Search

2: S ← some initial random candidate solution

Definition 1 Common Terms Used in Evolutionary Computation

Evolutionary Computation techniques are generally resampling techniques: new samples

Algorithm 17 An Abstract Generational Evolutionary Algorithm (EA)

Algorithm 18 The (µ, λ) Evolution Strategy

Algorithm 19 The (µ + λ) Evolution Strategy

Individuals Selected To Breed

Individuals Not Selected

(4, 8) Evolution Strategy

(1, 8) Evolution Strategy

3.1.1 Mutation and Evolutionary Programming

3.2 The Genetic Algorithm

University of Michigan Press.

Algorithm 21 Generate a Random Bit-Vector

3.2.1 Crossover and Mutation

Algorithm 22 Bit-Flip Mutation

Crossover is the Genetic Algorithm’s distinguishing

There are three classic ways of doing crossover in vectors:

Algorithm 23 One-Point Crossover

3: c ← random integer chosen uniformly from 1 to l inclusive