Machine Learning 9

Definition of Learning

A computer program is said to learn from experience E with respect to

some class of tasks T and performance measure P, if its performance at

tasks in T, as measured by P, improves with experience E.

Handwriting recognition

T: recognizing/classifying handwritten characters

E: handwritten characters with correct answers

P: percent of characters correctly recognized/classified

Vehicle driving

T: driving on public highways

E: front/sides/back view images and steering commands

P: distance traveled before an accident

Machine Learning 10

Issues in Designing Learning Systems

Type of experience (training examples)

Type of knowledge (hypotheses) to learn

Type of learning algorithms

Checkers playing

T: playing checkers

E: playing against people

P: chance of win

Machine Learning 11

Type of Experience

Direct or indirect feedback from teachers

Correct move for each board state vs. final outcome

Indirect feedback requires additional inference for credit assignment

Learning with direct feedback is easier, but it is usually not available

Active or passive control of training examples

Passive : teachers provide training examples (at random)

Active : learners ask for interesting examples

Explore : learners explore environments

Self learning : active & indirect learning (no teacher)

Quality of training examples

Difference between training examples and testing examples

Theoretical error bounds

Machine Learning 12

Type of Knowledge to Learn

Vehicle driving

Relationship between view images and control commands

Checkers playing

Relationship between the board states and best moves

• Target functions : B M

• Evaluation functions V: B

– Higher values, better board states

– V(b) = { +100, –100, 0, V(b) }

– V(b) = w

0

+ w

1

x

1

+ w

2

x

2

+ w

3

x

3

+ w

4

x

4

+ w

5

x

5

+ w

6

x

6

Machine Learning 13

Training Examples

A training example is an ordered pair

b, V

t

(b)

x

1

=3, x

2

=0, x

3

=1, x

4

=0, x

5

=0, x

6

=0, +100

x

1

=3, x

2

=3, x

3

=1, x

4

=0, x

5

=0, x

6

=0, ?

x

1

=3, x

2

=3, x

3

=0, x

4

=0, x

5

=0, x

6

=1, ?

How to generate training examples for indirect training experience

V

t

(b) V(S(b))

b, V(S(b))

Machine Learning 14

Training Algorithms

Learning find w

i

’s that best fit training examples

Minimize E = (V

t

(b) – V(b))

2

V(b) = w

0

+ w

1

x

1

+ w

2

x

2

+ w

3

x

3

+ w

4

x

4

+ w

5

x

5

+ w

6

x

6

Least mean squares (LMS) algorithm

while b (V

t

(b)-V’(b))0

for each training example b,V

t

(b)

compute V’(b) using current w

i

for each w

i

w

i

w

i

+ (V

t

(b)- V’(b))x

i

V

t

(b) – V(b) = + w

i

V

t

(b) – V(b) = – w

i

x

i

= 0 w

i

not changed

Machine Learning 15

Trace of the LSM algorithm

V(b) = w

0

+ w

1

x

1

+ w

2

x

2

+ w

3

x

3

+ w

4

x

4

+ w

5

x

5

+ w

6

x

6

w

i

w

i

+ (V

t

(b)- V’(b))x

i

x

1

=3, x

2

=1, x

3

=0, x

4

=0, x

5

=0, x

6

=0, +75 w

1

x

1

=1, x

2

=3, x

3

=0, x

4

=0, x

5

=0, x

6

=0, –75 w

2

Machine Learning 16

A Checkers Playing System

T: playing checkers

P: percent of games won in the world tournament

E: playing against itself

Target function (knowledge)

V: B

V(b) = w

0

+ w

1

x

1

+ w

2

x

2

+ w

3

x

3

+ w

4

x

4

+ w

5

x

5

+ w

6

x

6

Training examples (experience)

b, V

t

(b)

V

t

(b) V(S(b))

Training algorithm

LMS

Machine Learning 17

Components in Learning System – 1

Critic

Generates training examples

e.g. Trace of game {b, V

t

(b), … }

Generalizer

Learns from the training examples

e.g. Estimates the target function, i.e., w

i

’s

Experiment generator

Find new experience

Performance system

Performs the given task

e.g. Checkers playing program

Machine Learning 18

Components in Learning System – 2

Machine Learning 19

Conclusions

What types of training examples are available?

What (approximation) functions to learn?

How to represent the hypotheses?

What algorithms to use to learn?

How well the algorithms will perform?

How much training data is needed?

Machine Learning 20

Preview – 1

Chapter 2

Concept learning / inductive learning

Version spaces

Candidate-Elimination learning algorithm

Inductive bias

Task : Enjoy sport?

Chapter 3

Information theory : entropy

ID3 decision tree learning algorithm

Occam’s razor

Task : Enjoy sport?

Machine Learning 21

Preview – 2

Chapter 4

Neuron / perpectron

MLP (multi-layer perceptron) : backpropagation algorithm

Inductive bias

Task : vehicle driving, face recognition

Chapter 5

Statistics

Confidence intervals

Comparing learning algorithms

Machine Learning 22

Preview – 3

Chapter 6

Bayes theorem

Bayes optimal classifier learning algorithm

Naive Bayes classifier learning algorithm

Bayesian belief networks learning algorithm

EM (expectation maximization) algorithm

Comparison with CE/ANN/DT and Bayesian learning

Task : Text classification

Chapter 7

PAC (probably approximately correct) learning model

VC (Vapnik-Chervonenkis) dimension

Halving algorithm

Weighted-majority algorithm

Machine Learning 23

Assignments

Programming assignments

ID3

Error backpropagationn

Nave Bayes classifier

Task : Speech recognition (b/d/g/k/p/t classification)

Experiments

Presentations

## Comments 0

Log in to post a comment