0% found this document useful (0 votes)
42 views49 pages

IE506 IntrotoML 2024jan5

The document provides an overview of machine learning, including common applications and task types. It discusses supervised, unsupervised, and other learning techniques and gives examples like classification, clustering, and language processing. It also outlines different areas of machine learning like transfer learning and auto ML.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
42 views49 pages

IE506 IntrotoML 2024jan5

The document provides an overview of machine learning, including common applications and task types. It discusses supervised, unsupervised, and other learning techniques and gives examples like classification, clustering, and language processing. It also outlines different areas of machine learning like transfer learning and auto ML.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 49

A Broad Overview of Machine Learning and

Applications

IE 506
Lecture 0

January 5 & 9, 2024.

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 1 / 49
Outline

1 Introduction

2 Machine Learning - Motivating Applications

3 Nature of Machine Learning Tasks


Supervised Machine Learning
Classification Algorithms

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 2 / 49
Introduction

What is Machine Learning?

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 3 / 49
Machine Learning - Motivating Applications

Machine Learning - Motivating Applications

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 4 / 49
Machine Learning - Motivating Applications

Binary Classification
e-mail Spam Classification

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 5 / 49
Machine Learning - Motivating Applications

Multi-class Classification
Handwriting Recognition

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 6 / 49
Machine Learning - Motivating Applications

Multi-class Classification
Handwriting Recognition

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 7 / 49
Machine Learning - Motivating Applications

Multi-class Classification on Video Data

Action Recognition

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 8 / 49
Machine Learning - Motivating Applications

Multi-label Classification
Object Recognition

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 9 / 49
Machine Learning - Motivating Applications

Multi-task and Multi-label Classification


Object Recognition and Localization

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 10 / 49
Machine Learning - Motivating Applications

Multi-task and Multi-label Classification


Object Recognition and Localization

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 11 / 49
Machine Learning - Motivating Applications

Clustering

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 12 / 49
Machine Learning - Motivating Applications

Principal Component Analysis

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 13 / 49
Machine Learning - Motivating Applications

Natural Language Processing Applications

Part-of-Speech Tagging

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 14 / 49
Machine Learning - Motivating Applications

Natural Language Processing Applications


Parse Tree Prediction

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 15 / 49
Machine Learning - Motivating Applications

Natural Language Processing Applications

Machine Translation

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 16 / 49
Machine Learning - Motivating Applications

Computer Vision Applications


Image Segmentation

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 17 / 49
Machine Learning - Motivating Applications

Computer Vision Applications


Image Generation

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 18 / 49
Machine Learning - Motivating Applications

Computational Biology
Protein Structure Prediction

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 19 / 49
Nature of Machine Learning Tasks

Nature of Machine Learning Tasks

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 20 / 49
Nature of Machine Learning Tasks

Nature of Machine Learning Tasks

Supervised Learning
▶ Inputs and corresponding outputs are known during learning
▶ e.g. Regression, Classification (Binary, Multi-class, Multi-label)
Unsupervised Learning
▶ Input objects are generally not labeled
▶ e.g. Clustering, Principal-component Analysis
Semi-supervised Learning
▶ learning from a few labeled data
▶ e.g. customer reviews are available only for a few products in a seller’s
website

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 21 / 49
Nature of Machine Learning Tasks

Nature of Machine Learning Tasks

Weakly supervised Learning


▶ labels for inputs are generally unspecific and sparse
▶ e.g. when precise localization of objects is to be done based on only
coarse labels.
Online learning
▶ typically used in context of a supervised learning setup
▶ the entire data set is not stored or available for training
▶ training of ML algorithm done using input, output pairs arriving in
sequential fashion
▶ model update happens in a purely incremental fashion
Reinforcement Learning
▶ learning to choose an action at a system’s state in the presence of an
environmental signal so that some objective is met
▶ e.g. learn to navigate in a crowded street

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 22 / 49
Nature of Machine Learning Tasks

Nature of Machine Learning Tasks


Transfer Learning
▶ transferring a learned model from task T1 to T2
▶ e.g. transfer from image captioning to video captioning
Auto ML
▶ automating data pre-processing
▶ automating feature generation
▶ automating selection of ML algorithms for a particular task
▶ automating selection of hyperparameters in an ML algorithm
▶ automating post-processing results obtained from ML algorithms
▶ automating analysis and interpretation of results
▶ e.g. Architecture search in neural networks
Meta-learning
▶ learning from meta-data obtained from ML algorithms
▶ e.g. understanding which connections in a neural network lead to a
particular prediction

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 23 / 49
Nature of Machine Learning Tasks

Nature of Machine Learning Tasks

Several other learning buzzwords in use


▶ self-supervised learning
▶ lifelong learning
▶ curriculum learning
▶ mutual learning
▶ one-shot/few-shot learning
▶ knowledge distillation (or) teacher-student model based learning
▶ and others...

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 24 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Supervised Machine Learning

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 25 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Binary Classification
Recall: e-mail Spam Classification

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 26 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Binary Classification

Input: e-mail messages


Output: Spam/Not spam

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 27 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Binary Classification

Input: e-mail messages =⇒ some feature space


Output: Spam/Not spam =⇒ {+1, −1}

Generally many input/output pairs are given for learning the machine
learning model.

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 28 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Binary Classification
Feature Extraction

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 29 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Binary Classification
Feature Extraction

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 30 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Binary Classification

Input: e-mail messages =⇒ some feature space ⊆ Rd .


▶ x ∈ X ⊆ Rd

Output: Spam/Not spam =⇒ {+1, −1}


▶ y ∈ Y = {+1, −1}

Generally n input/output pairs {(x i , y i )}ni=1 ∈ (X × Y)n are given for


learning the machine learning model.

D = {(x i , y i )}ni=1 called the training data.

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 31 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

A sample data set for email classification

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 32 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Representing image as vectors

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 33 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

General Nature of a Supervised Machine Learning Task

Training
Input: Training data D = {(x i , y i )}ni=1
Aim: Learn a model h : X → Y

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 34 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

General Nature of a Supervised Machine Learning Task

Training
Input: Training data D = {(x i , y i )}ni=1
Aim: Learn a model h : X → Y
Testing
Given x̂, predict ŷ = h(x̂)
P. Balamurugan A Broad Overview of Machine Learning and Applications
January 5 & 9, 2024. 35 / 49
Nature of Machine Learning Tasks Supervised Machine Learning

Types of Machine Learning Algorithms for Classification

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 36 / 49
Nature of Machine Learning Tasks Classification Algorithms

Classification Algorithms

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 37 / 49
Nature of Machine Learning Tasks Classification Algorithms

SVM

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 38 / 49
Nature of Machine Learning Tasks Classification Algorithms

SVM

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 39 / 49
Nature of Machine Learning Tasks Classification Algorithms

Decision Tree

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 40 / 49
Nature of Machine Learning Tasks Classification Algorithms

Decision Tree

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 41 / 49
Nature of Machine Learning Tasks Classification Algorithms

Regression

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 42 / 49
Nature of Machine Learning Tasks Classification Algorithms

Regression

Fitting a curve

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 43 / 49
Nature of Machine Learning Tasks Classification Algorithms

Regression

Fitting a curve

Pn
Model: y (x; β) = β0 + j=1 βj ϕj (x)

{ϕ1 (x), ϕ2 (x), . . . , ϕn (x)} represents set of basis functions.

β0 , β1 , . . . , βn represent model parameters.

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 44 / 49
Nature of Machine Learning Tasks Classification Algorithms

Regression
Fitting a curve

Pn
Model: y (x; β) = j=0 βj ϕj (x)

{ϕ0 (x), ϕ1 (x), ϕ2 (x), . . . , ϕn (x)} represents set of basis functions.


ϕ0 (x) = 1 can be assumed.

β0 , β1 , . . . , βn represent model parameters.

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 45 / 49
Nature of Machine Learning Tasks Classification Algorithms

Regression
Fitting a curve

Pn
Model: y (x; β) = β0 + j=1 βj ϕj (x)
Expressivity and Dimensionality:
▶ For f1 , the basis functions are {ϕ0 (x) = 1, ϕ1 (x) = x, ϕ2 (x) = x 2 }.
▶ For f2 , the basis functions are {ϕ0 (x) = 1, ϕ1 (x) = x, ϕ2 (x) =
x 2 , ϕ3 (x) = x 3 , ϕ4 (x) = x 4 , ϕ5 (x) = x 5 }.
▶ f1 is low-dimensional (only three basis functions) but less expressive.
▶ f2 is high-dimensional (six basis functions) and better expressive than
f1 .
P. Balamurugan A Broad Overview of Machine Learning and Applications
January 5 & 9, 2024. 46 / 49
Nature of Machine Learning Tasks Classification Algorithms

Regression

Fitting a curve

Pn
Model: y (x; β) = β0 + j=1 βj ϕj (x)

Consequences of Expressivity and Dimensionality:


▶ Curse of dimensionality, Occam’s razor principle, bias-variance tradeoff,
overfitting, and many more.

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 47 / 49
Nature of Machine Learning Tasks Classification Algorithms

Data Exploration

Homework:
Check the UCI data repository, https://archive.ics.uci.edu/.
Explore at least 5 different data sets. Perform the following
▶ Understand the number, type and description of features or attributes.
▶ Understand the number and nature of samples.
▶ Understand how the data was acquired, check if there are missing data,
etc.
▶ Check for any other relevant qualities of the data.
Write Python code to load these 5 data sets and create Pandas data
frames from the data loaded.

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 48 / 49
Nature of Machine Learning Tasks Classification Algorithms

ACKNOWLEDGMENTS
Some content borrowed from various open-access resources
▶ Blogs
▶ Tutorials
▶ Free e-books
▶ Open-access Papers
▶ Youtube videos
▶ Scribe notes from my students

P. Balamurugan A Broad Overview of Machine Learning and Applications


January 5 & 9, 2024. 49 / 49

You might also like