What This Chapter Is About

- Learning and Concepts

- Why Machine Learning

- Types of Learning

- Neural Networks

- Genetic Algorithm

- Fuzzy Learning

- Boltzman Machines

Machine Learning

One of ways to achieve AI

AI Hierarchy

Learning

What is Learning

- Acquiring new knowledge

- Modifying existing knowledge

How

- Changes

- Generalization

- Improvement

Different Learning Methods

Inductive Learning

Deductive Learning

Inductive Learning

Find patterns from data

Deductive Learning

Derive conclusion from facts

A=B,B=C -> A=C

Learning

Rote Learning

Memorization

Rote Learning

Learning by Analogy

Explanation-Based Learning

Learning by observing and analyzing human solutions to specific problems

After lerning, go directly from facts to solution

Prototypical EBL Architecture

Explanation-Based Learning

- Needs many examples

- Broken Explanations?

Learning from Examples

- Inductive Learning

Inductive Learning

Inductive Learning

- Find Patterns

Learning Framework

Environment

- Nature and Quality of Information

Information Level

- High Level -> Abstract -> Broad Class of Problems

- Low Level -> Details -> Single Problem

Information Quality

- Noise Free

- Reliable

- Ordered

Knowledge Base

- Expressive

- Modifiable

- Extensibility

Genetic Algorithm

Modeling Search as Evolution

- Crossover

- Mutation

- Survival of the fittest

- Survival of the most diverse

Rocket Evolution

Simple Evolving Organism

- Heredity

- Variation

- Selection

General Algorithm

A) generate initial population

B) evaluate fitness of all

C) select fitter population for crossover

D) cross-over

E) mutate

F) evaluate fitness of offspring

Loop while reaching desired fitness

Roulette Wheel

Disadvantage

- Less efficient in terms of speed of convergence

- Tendency to get stuck in local maxima

Fuzzy Learning

Crisp

Fuzzy

Fuzzy Rule-Based System

Fuzzy Logic Operators

Fuzzy Logic Operators

De-Fuzzify Result