GENETIC ALGORITHMS

1
GENETIC ALGORITHMS

• IS3301 Session 19b
• Prof. Mark Nissen

2
Agenda

• Machine Learning
• ACME Transportation Problem
• Genetic Algorithms
• Learning Exercise
• Examples Demos
• Summary

3
Machine Learning

• Increase knowledge in systems
• Generalize across specific instances
• Knowledge enhances problem solving
• Learning enhances knowledge
• Rule-based problems
• Expert not always available/willing/able
• Knowledge engineering bottleneck
• No one knows how to solve problem
• Static knowledgebase, dynamic environ

4
Machine Learning Approaches

• Symbolic
• Induction - rules from examples (ID3/VPX)
• Chunking - generalize rules (Soar)
• Analogy - reason from past cases (CBR)
• Explanation - generalize with theory (EBL)
• Numerical
• Statistics - fit models to data (regression)
• Neural nets - fit network to data (ANN)
• Genetic algorithms - evolve pop to fit data
• Supervised training most common

5
AMCE Transport Problem

• Plan delivery routes
• 25 overseas locations
• 25! Different possible routes
• Evaluate 1M routes/second
• 0.25 of routes since life began on Earth
• NP-complete problem
• Computationally intractable
• Exhaustive search impractical

6
AMCE Transport Problem

• Optimization difficulties?
• Nonlinear functions constraints
• Discontinuities
• Large, sparse search space
• Qualitative factors
• Dont know how to solve problem
• Simulation difficulties?
• Expert system difficulties?
• Gulf War Syndrome example

7
Genetic Algorithms

• Biological metaphor (Darwinian)
• Survival of the fittest (problem solver)
• Evolve population of problem solvers
• Computational chromosomes
• 3 operators
• Selection - only fittest survive
• Crossbreeding - exchange genes
• Mutation - random variation
• Base on fitness in problem space
• Population learns as it evolves

8
Learning Process
Random Population Evolved
Population 1
YZA
MNO
ABC
DEF
PQR
. . . . . . .
STU
GHI
JKL
VWX
Environmental evaluation
9
Mechanics

• ID chromosomes (feature vectors)
• Capture key problem characteristics
• Symbols or numbers (often binary)
• Randomly populate with diverse genes
• Establish fitness function
• Evaluate performance in environment
• Apply operators to population
• Evolve population through generations

10
Learning Exercise

• English vowels (A,E,I,O,U)
• Binary grid representation (0,1)
• Start with 5 random vectors (9-bit)
• Evaluation correct bits
• Select 2 vectors with highest score
• Crossbreed 1 (4 bits from a, 5 from b)
• Mutate best vector (keep 6 bits, change 3)
• Insert 1 totally-random vector
• Evaluate evolve until letter is learned

11
Examples Demos

• Computational examples
• i85l19b1.xls
• i85l19b2.xls
• Curve-fitting demo (check)
• Other examples
• Other demos
• Naval Research Lab
• John Koza (Stanford) video
• Others for you to discover

12
Summary

• Learning - d/dx (knowledge)
• GAs are powerful
• NP-complete problems
• Difficulties with optimization, others
• Dont have to know how to solve problem!
• GAs use biological metaphor
• Darwinian level
• Selection, crossover mutation ops
• Evolve population of problem solvers