Incremental Learning

ITS69404 Cognitive Computing and
applications
School of Computer Science
Taylor’s University Lakeside Campus
Lecture 7: Incremental Learning

Learning Outcome
• At the end of this topic, you should be able to:

• Discuss the concept of Incremental learning and
hence, understand analogical reasoning that is
a core process of cognition.
• Aligns to Module Learning Outcome 1
Lecture 7 - Incremental Learning 2

Outline
• Purpose of incremental learning

• Variabilization
• Specialization
• Generalization
• Heuristics for specialization and
generalization

Introduction
Learning by
Storing Cases
Incremental
Learning
IN CASE-BASED REASONING WE HOW CAN WE ABSTRACT CONCEPTS

STORED NEW EXAMPLES IN MEMORY FROM THOSE EXAMPLES?

Problem on Incremental Learning

Problem on • Background knowledge plays an
Incremental important role in deciding whether this
Learning contd. example is a foo!

Problem on
Incremental Learning
contd.
• Again, background
knowledge plays an
important role in
deciding whether this
example is a foo or
not!

Problem on
Incrementa
l Learning
contd.

lessons that could be learnt from this
problem
1) Learning is often incremental
• We learn from one example at a time
• Cognitive systems, intelligent systems or human beings are not always given
hundreds or thousands or millions of examples all at a time. Only one example at a
time
2) Often the examples are labelled
• It is the teacher who tells which is a positive example and which is negative. This is
called supervised learning in machine learning
3) The examples can come in a particular order
• There are some positive examples and some negative examples
4) In case-based reasoning the examples were stored in raw form in the memory, and
they were reused
• In incremental learning, concept is abstracted
5) The number of examples from which concepts are abstracted is very small
6) What exactly to abstract or learn or generalize becomes a hard problem
• There is often a tendency to over specialize or over generalize
• How the intelligent agent finds out exactly what is the right level of
generalization?
Basic algorithm
for incremental
learning

Incremental
learning more
systematically
• Teaching AI
algorithm the
concept of an arch
11
Lecture 7 - Incremental Learning
Incremental
learning more
systematically
contd.
• 2nd example

Heuristics for Generalization and
Specialization
• Drop link heuristics. Heuristics

is a rule of thumb

Heuristics for Generalization and
Specialization contd.
• Negative example of an arch.

Heuristics for
Generalization
and Specialization
contd.
• Require-link
heuristics

Heuristics for Generalization and Specialization
contd.
• 3rd example

contd.
• Specializing the negative example
• Forbid-link heuristics

contd.
• 4th example

contd.
• Enlarge-set heuristics

contd.
• Climb-tree heuristics
• Use the background knowledge to generalize

contd.
• Summary of heuristics that an AI agent may use for the concept of
incremental learning

Variabilization
• The concept the AI

agent is going to learn
is called a foo
• 1st example given to the
AI algorithm
• What will be the values
the boxes in the right
will take

22
Variabilization
contd.
• Initially the bricks
were labelled
brick 1, 2, 3 etc.,
however we can
abstract them as
instance of bricks

Variabilization
contd.
• How to represent
the Relationships
of the concept on
the right?

25
Refining the
concept
definition..

• 2nd example given to
the AI programme
• How to refine this to
exclude the negative
example?
26
Refining the
concept
definition..

• answer
Refining the
concept
definition..
• Another positive
example
• How to refine the
current concept?

Refining the
concept
definition..

Refining the concept definition..
• Suppose the AI agent has background knowledge that Brick, and Cylinder are both
sub-concepts of a Block
• How does the AI program refine the current concept definition?

• Knowledge plays a critical role in guiding the learning process

• What we learn at the end depends on what background knowledge we begin with

• A negative example given to the AI program
• How the AI agent may refine this concept to exclude this negative
example?
• The example is a negative instance and the current definition of

the concept already excludes it then we don’t have to do anything
Refining the concept
definition..
• Given the input with series
of example and background
knowledge, below is the
final concept of a foo that
the AI agent will learn
• The result of learning here
depends not just on the
input examples, but on the
background knowledge that
the AI agent has

Purpose of Incremental Concept Learning
• This method of incremental concept of learning differs from some of the standard ML
algorithms
• Often in ML, the AI agent is given a large number of examples to begin the learning
with
• Then we can apply statistical methods to find patterns of regularity in the input data
• If the examples are small and come one at a time, then the learning is incremental
• Then it becomes challenging to apply those statistical methods to detect patterns of
regularity in the input data
• In that case the algorithm must make use of its background knowledge to decide what
to learn and how to learn

Conclusion
• Purpose of incremental concept learning
• When we have limited number of examples then we need to learn concept
based on these limited examples
• Variabilization
• Deriving more abstract notion of concept
• Specialization or generalization
• Based on positive or negative examples we generalized or specialized our
understanding
• Heuristics for specialization and generalization
• Forbid-link or require-link that allows to develop better concept based on
examples
• Incremental concept of learning is similar to and connected with human cognition

Q & A!

Incremental Learning

Uploaded by

Copyright:

Available Formats

Incremental Learning

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

Incremental Learning

Uploaded by

Copyright:

Available Formats

ITS69404 Cognitive Computing and

Lecture 7: Incremental Learning

• At the end of this topic, you should be able to:

• Aligns to Module Learning Outcome 1

Lecture 7 - Incremental Learning 2

• Purpose of incremental learning

Lecture 7 - Incremental Learning 3

IN CASE-BASED REASONING WE HOW CAN WE ABSTRACT CONCEPTS

Lecture 7 - Incremental Learning 4

Lecture 7 - Incremental Learning 5

Lecture 7 - Incremental Learning 6

Lecture 7 - Incremental Learning 7

Lecture 7 - Incremental Learning 8

Lecture 7 - Incremental Learning 10

Lecture 7 - Incremental Learning 12

• Drop link heuristics. Heuristics

Lecture 7 - Incremental Learning 13

• Negative example of an arch.

Lecture 7 - Incremental Learning 14

Lecture 7 - Incremental Learning 15

Lecture 7 - Incremental Learning 16

Lecture 7 - Incremental Learning 17

Lecture 7 - Incremental Learning 18

Lecture 7 - Incremental Learning 19

• Use the background knowledge to generalize

Lecture 7 - Incremental Learning 20

Lecture 7 - Incremental Learning 21

• The concept the AI

Lecture 7 - Incremental Learning

Lecture 7 - Incremental Learning 23

Lecture 7 - Incremental Learning 24

Lecture 7 - Incremental Learning

Lecture 7 - Incremental Learning

Lecture 7 - Incremental Learning 27

Lecture 7 - Incremental Learning 28

Lecture 7 - Incremental Learning 29

• Knowledge plays a critical role in guiding the learning process

Lecture 7 - Incremental Learning 30

• The example is a negative instance and the current definition of

Lecture 7 - Incremental Learning 32

Lecture 7 - Incremental Learning 33

Lecture 7 - Incremental Learning 34

You might also like