1. 진화방식을 이용한 인공시스템의 기능 및 구조의 창발
조 성 배
연세대학교 컴퓨터과학과

2. Evolutionary Computation
Different lines of investigation in simulated evolution
GA (Genetic Algorithms)
ES (Evolution Strategies)
EP (Evolutionary Programming)
Each abstracts evolution in a slightly different manner

3. Evolutionary Theory for Simulation
Neo-darwinian paradigm :
Synthesis of classic darwinism
selectionism of Weismann
genetics of Mendel
Statistical processes within population
Reproduction
Mutation
Competition
Selection

4. Duality of Living Organism
Genotype : underlying genetic coding
Phenotype : manner of response contained in physiology, morphology
and behavior of organisms
Phenotypic
state space p1 p2
p1’
g1’ g2 g1 G P f1 f2 f3 f4
f1’
Epigenesis :
f1 : I  G  P
Selection :
f2 : P  P
Genotypic survival :
f3 : P  G
Mutation :
f4 : G  G
Genotypic
state space

5. Overview of EC Similarity Population-based search
Impose random variation
Apply selection operation
Difference
Typical representation
Emphasis on type of variation imposed
Promoting vs. culling view of selection
Three main lines of investigation
GA, ES, EP EA EC

6. Sources Genetic Algorithms
First book in field : John Holland, Adaptation in Natural and Artificial Systems, MIT Press, 1991.
Theory-oriented textbook : David E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning, Addison-Wesley, 1989.
Application-oriented textbook : Lawrence Davis, Handbook of Genetic Algorithms, Van Nostrand Reinhold, 1991.

7. Overview Genetic Algorithms Invented by John Holland (early 1970s)
Techniques for optimization and machine learning inspired by features of theory of biological evolution
Solution chromosomes
Search proceeds through maintenance of a population of solutions
Reproduction biased in favor of better individuals
New encodings created by mutation and recombination
encoded

8. Five Components Genetic Algorithms
Encoding technique (“chromosome structure”)
Evaluation function (“environment”)
Initialization procedure (“creation”)
Genetic operators (mutation, recombination, etc)
Parameter settings (practice and art)

9. Procedure Genetic Algorithms reproduction discard population
evaluation
modification
parents
children
deleted
members
modified
evaluated

10. Population of Chromosomes
Genetic Algorithms
Population of Chromosomes
Bit string ( … )
Real number lists ( … )
Permutation of elements ( E26 E13 E18 … E4 E9 )
etc

11. Reproduction Genetic Algorithms
Parents are selected at random, with selection chances biased in relation to chromosome evaluations
ex) parent selection (expected children = average / evaluation)
average = 10
cf) Roulette Wheel Parent Selection
Given parent 1 with evaluation 10
Chances of being chosen are
proportional to evaluation 3

12. Chromosome Modification
Genetic Algorithms
Chromosome Modification
Triggered stochastically
Operators : mutation / crossover (recombination)
(1) mutation causes a local modification
( ) ( )
( ) ( )
(2) crossover causes recombination
( ) ( )
( ) ( )
important feature of GA
accelerate search early in evolution of a population
lead to effective combination of schemata (subsolutions)

13. Evaluation and Deletion
Genetic Algorithms
Evaluation and Deletion
Evaluation
evaluator decodes a chromosome and assigns it a number measuring its success
the only link between classical GA and problem
Deletion
generational approach : entire population can be replaced at a time
steady-state approach : a few members can be replaced at a time

14. An Example (1) Genetic Algorithms f(x) = x2 where 0  x  31 x f(x) 3131
decoding
5-bit unsigned binary integer
ex) 13 = 01101
initial population (at random)
population size = 4
tossing a fair coin 20 times

15. An Example (2) Genetic Algorithms 1 2 3 4 sum average max Population
x value 13 24 8 19 x2
f(x) value
169
576 64
361
1170
293
fi / f
Pselecti
0.14
0.49
0.06
0.31
1.00
0.25
fi / f
expected
count
0.58
1.97
0.22
1.23
4.00
mate
crossover x 12 25 27 16
f(x)
1754
439
729

16. Schema Genetic Algorithms Combination of chromosome features
ex) ( ) contains schemata
(# 1 1 # #), (# 1 # 0 #), (0 1 1 # 1), ( ), (# # # # #)
Holland’s schema theorem
Nt : no. of occurrences S at t
Et : avg. evaluation of S at t
At : avg. evaluation of entire population at t
better than avg. : proliferate
bad ones : die out

17. Hybrid GA Genetic Algorithms
Combination of GA and one or more other optimization algorithms
If done correctly, it can’t do worse
Combining search capability of GA with domain-based expertise

18. Creating a Hybrid GA Genetic Algorithms decoder domain knowledge
representation
operators
population
initialization
domain algorithm
domain knowledge

19. Applications Genetic Algorithms
Control : gas pipeline, pole balancing, missile pursuit
Scheduling : production / facility
Game playing : poker, prisoner’s dilemma
Design : semiconductor layout, aircraft design, communication network
design, keyboard configuration
Combinatorial optimization : set covering, TSP, routing,
graph coloring / partitioning
Machine learning : designing NN, improving classification algorithms,
classifier systems
Signal processing : filter design
Robotics : trajectory planning

20. Some Current Efforts Genetic Algorithms
Analyzing mathematical properties
Artificial life simulations
Practical real world applications
Parallel distributed processing

21. 연구 분야 지능형 이동 로봇 에이전트 제어 기반기술 MNN CAM-Brain Fuzzy-GA Fuzzy NN IGA
A-Life
이동 로봇 제어
패턴인식/
미디어검색
지능형
에이전트 자동
프로그래밍
네트워크 게임
기반기술