SEN 503/339 Artificial Intelligence

Machine Learning Fundamentals-I

Lec 5.1: Introduction to Machine Learning

Halûk Gümüşkaya
Professor of Computer Engineering

web: [Link]
e-mail: haluk@[Link], halukgumuskaya@[Link]

: [Link]
: [Link]

Introduction to Machine Learning

1. What is Machine Learning?
2. Machine Learning Knowledge Areas
3. Types of Learning
4. Supervised Learning
5. Unsupervised Learning
6. Semi-Supervised Learning
7. Reinforcement Learning

Haluk Gümüşkaya @ [Link] 2

 Some Definitions for Machine Learning (1/2)
[Arthur Samuel, 1959]
 Field of study that gives computers the ability to learn without
being explicitly programmed .

[Kupfermann 1985]
 Learning is the acquisition of knowledge about the world.

[Shepherd 1988]
 Learning is an adaptive change in behavior caused by
experience.

Haluk Gümüşkaya @ [Link] 3

Some Definitions for Machine Learning (2/2)

[Kevin Murphy] algorithms that
 automatically detect patterns in data
 use the uncovered patterns to predict future data or other
outcomes of interest

[Tom Mitchell] algorithms that

 improve their performance (P)
 at some task (T)
 With experience (E)

Haluk Gümüşkaya @ [Link] 4

Comparison: ML vs Traditional Programming
 Traditional Programming

 Machine Learning

Haluk Gümüşkaya @ [Link] 5

Empirical Inference
 Drawing conclusions or finding a model that represent data
samples from empirical data (observations, measurements).
 Example 1: Scientific inference

Haluk Gümüşkaya @ [Link] 6

 Empirical Inference
 Drawing conclusions or finding a model that represent data
samples from empirical data (observations, measurements).
 Example 1: Scientific inference

Haluk Gümüşkaya @ [Link] 7

Empirical Inference
 Drawing conclusions or finding a model that represent data
samples from empirical data (observations, measurements).

 Example 1: Perception

“The brain is nothing but a statistical decision organ.”

H. Barlow

Haluk Gümüşkaya @ [Link] 8

 Perception

Haluk Gümüşkaya @ [Link] 9

Hard Inference Problems

 High dimensionality
 consider many factors simultaneously to find regularity

 Complex regularities
 nonlinear; nonstationary, etc.

 Little prior knowledge

 e.g. no mechanistic models for the data

 Need large data sets

 processing requires computers and automatic inference
methods
Haluk Gümüşkaya @ [Link] 10
 Why Study Machine Learning?
Engineering Better Computing Systems

Develop systems
• too difficult/expensive to construct manually
• because they require specific detailed skills/knowledge
• knowledge engineering bottleneck
• that adapt and customize themselves to individual users.
• Personalized news or mail filter
• Personalized tutoring
•…

Discover new knowledge from large databases

• Medical text mining
(e.g. migraines to calcium channel blockers to magnesium)
• data mining
Haluk Gümüşkaya @ [Link] 11

Why Study Machine Learning?

Cognitive Science

Computational studies of learning

• may help us understand learning in humans
• and other biological organisms.

Hebbian neural learning

• “Neurons that fire together, wire together.”

Haluk Gümüşkaya @ [Link] 12

 Why are things working today?
Algorithms
 Many basic effective and
efficient algorithms available.

Data
• Large amounts of on-line
data available.

Computing
• Large amounts of
computational resources
available

Haluk Gümüşkaya @ [Link] 13

Introduction to Machine Learning

1. What is Machine Learning?
2. Machine Learning Knowledge Areas
3. Types of Learning
4. Supervised Learning
5. Unsupervised Learning
6. Semi-Supervised Learning
7. Reinforcement Learning

Haluk Gümüşkaya @ [Link] 14

 Where does ML fit in?

Haluk Gümüşkaya @ [Link] 15

Machine Learning Knowledge Areas

Haluk Gümüşkaya @ [Link] 16

 Machine Learning Algorithms

Haluk Gümüşkaya @ [Link] 17

Machine Learning in One Picture

Types of Learning:
Simple
Classification

Haluk Gümüşkaya @ [Link] 18

 Introduction to Machine Learning
1. What is Machine Learning?
2. Machine Learning Knowledge Areas
3. Types of Learning
4. Supervised Learning
5. Unsupervised Learning
6. Semi-Supervised Learning
7. Reinforcement Learning

Haluk Gümüşkaya @ [Link] 19

Classes of Learning Problems

Supervised Unsupervised Reinforcement

Learning Learning Learning

Teach the computer how to do Let the computer learn how to do

something, then let it use its new something, and use this to determine
found knowledge to do it. structure and patterns in data.
[Link]
Haluk Gümüşkaya @ [Link] 20
 3 Types of Machine Learning

Haluk Gümüşkaya @ [Link] 21

Machine Learning Summarized in One Picture

Haluk Gümüşkaya @ [Link] 22

 Types of Learning

Haluk Gümüşkaya @ [Link] 23

4 Types of Machine Learning

Haluk Gümüşkaya @ [Link] 24

 Types of Learning

Haluk Gümüşkaya @ [Link] 25

Classes of Learning Problems

Haluk Gümüşkaya @ [Link] 26

 Classes of Learning Problems: Comparison

[Link]
Haluk Gümüşkaya @ [Link] 27

Introduction to Machine Learning

1. What is Machine Learning?
2. Machine Learning Knowledge Areas
3. Types of Learning
4. Supervised Learning
5. Unsupervised Learning
6. Semi-Supervised Learning
7. Reinforcement Learning

Haluk Gümüşkaya @ [Link] 28

Supervised Learning

Haluk Gümüşkaya @ [Link] 29

Supervised Learning

Haluk Gümüşkaya @ [Link] 30

Supervised Learning

Haluk Gümüşkaya @ [Link] 31

Binary Classification

Haluk Gümüşkaya @ [Link] 32

Multiclass Classification + Annotation

Haluk Gümüşkaya @ [Link] 33

Regression

Haluk Gümüşkaya @ [Link] 34

Sequence Annotation

given sequence:
 gene finding
 speech recognition
 activity segmentation
 named entities

Haluk Gümüşkaya @ [Link] 35

Ontology

Haluk Gümüşkaya @ [Link] 36

 Prediction

tomorrow’s stock price

Haluk Gümüşkaya @ [Link] 37

Example 1: Supervised Learning

Classification on Tabular Data
Training Data: Golf

Haluk Gümüşkaya @ [Link] 38

 ML Example 1: Supervised Learning:
Classification using RapidMiner
Decision Tree Based Classification

Machine Learning: Data-Driven Approach

Haluk Gümüşkaya @ [Link] 39

Decision Tree Based Classification: Model Generation

Haluk Gümüşkaya @ [Link] 40

 ExampleSet (Retrieve Golf-Testset)

Haluk Gümüşkaya @ [Link] 41

Performance Vector: Accuracy and Confusion Matrix

Haluk Gümüşkaya @ [Link] 42

 ML Example 2: Image Classification
1. Collect a dataset of images and labels
2. Use Machine Learning to train an image classifier
3. Evaluate the classifier on a withheld set of test images

Haluk Gümüşkaya @ [Link] 43

First Classifier: Nearest Neighbor Classifier

Haluk Gümüşkaya @ [Link] 44

 Introduction to Machine Learning
1. What is Machine Learning?
2. Machine Learning Knowledge Areas
3. Types of Learning
4. Supervised Learning
5. Unsupervised Learning
6. Semi-Supervised Learning
7. Reinforcement Learning

Haluk Gümüşkaya @ [Link] 45

Unsupervised Learning

Haluk Gümüşkaya @ [Link] 46

 Unsupervised Learning

Haluk Gümüşkaya @ [Link] 47

Unsupervised Learning
 Given data x, ask a good question ... about x or about model for x

 Clustering
Find a set of prototypes representing the data

 Principal Components
Find a subspace representing the data

 Sequence Analysis
Find a latent causal sequence for observations
 Sequence Segmentation
 Hidden Markov Model (discrete state)
 Kalman Filter (continuous state)

 Hierarchical representations
 Independent components / dictionary learning
Find (small) set of factors for observation
Novelty detection: Find the odd one out
Haluk Gümüşkaya @ [Link] 48
 Clustering
 Documents
 Users
 Webpages
 Diseases
 Pictures
 Vehicles

Haluk Gümüşkaya @ [Link] 49

Principal Components

Variance component model to account for sample structure in genome-wide

association studies, Nature Genetics 2010
[Link]
Haluk Gümüşkaya @ [Link] 50
 Hierarchical Grouping

Haluk Gümüşkaya @ [Link] 51

Independent Components

find them
automatically

Haluk Gümüşkaya @ [Link] 52

 Novelty Detection

Haluk Gümüşkaya @ [Link] 53

Introduction to Machine Learning

1. What is Machine Learning?
2. Machine Learning Knowledge Areas
3. Types of Learning
4. Supervised Learning
5. Unsupervised Learning
6. Semi-Supervised Learning
7. Reinforcement Learning

Haluk Gümüşkaya @ [Link] 54

 Semi-Supervised Learning

Try to improve predictions based on examples by making use of

the additional “unlabeled” examples.
Haluk Gümüşkaya @ [Link] 55

Introduction to Machine Learning

1. What is Machine Learning?
2. Machine Learning Knowledge Areas
3. Types of Learning
4. Supervised Learning
5. Unsupervised Learning
6. Semi-Supervised Learning
7. Reinforcement Learning

Haluk Gümüşkaya @ [Link] 56

Supervised Learning vs Reinforcement Learning

Haluk Gümüşkaya @ [Link] 57

Machine Learning: Supervised vs Reinforcement

Haluk Gümüşkaya @ [Link] 58

 Reinforcement Learning in Humans

Haluk Gümüşkaya @ [Link] 59

Open Question: What can be Learned from Data?

Haluk Gümüşkaya @ [Link] 60

Self Driving Car: Environment, Sensors and Data

Haluk Gümüşkaya @ [Link] 61

Self Driving Car: Feature Extraction and Representation

Haluk Gümüşkaya @ [Link] 62

Self Driving Car: Representation to Knowledge

Haluk Gümüşkaya @ [Link] 63

Self Driving Car: Reasoning

Haluk Gümüşkaya @ [Link] 64

Self Driving Car: Planning Actions and Executing

Haluk Gümüşkaya @ [Link] 65

Deep Learning and Deep Reinforcement Learning

Haluk Gümüşkaya @ [Link] 66

 Important Challenges in ML
 How important is the actual learning algorithm and its tuning
 Simple versus complex algorithm
 Overfitting
 Model Selection
 Regularization
…

Haluk Gümüşkaya @ [Link] 67