21
7. Genetic Programming 7. Genetic Programming and Emergent Order and Emergent Order GP-Seminar 98.9.19 신신신

7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

Embed Size (px)

Citation preview

Page 1: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

7. Genetic Programming 7. Genetic Programming and Emergent Orderand Emergent Order

GP-Seminar

98.9.19 신수용

Page 2: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

2

ContentsContents

Introduction Evolution of Structure and Variable Length Genomes Iteration, Selection, and Variable Length Program Structures Evolvable Representations The Emergence of Introns, Junk DNA, and Bloat Introns in GP defined Why GP Introns Emerge Effective Fitness and Operator Why Introns Grow Exponentially The Effects of Introns What to Do about Introns

Page 3: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

3

7.1 Introduction7.1 Introduction

This chapter focuses on emergent properties arising from GP’s freedom of representation of the problem space

Two important properties of GP GP’s ability to search the space of the problem representatio

n The problem of introns or bloat

emergence -> variable length genotype 때문

Page 4: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

4

7.2 Evolution of Structure and 7.2 Evolution of Structure and Variable Length GenomesVariable Length Genomes

The capability to evolve a representation of the problem depends on the ability of a learning algorithm to modify the structure of its own solutions

By evolving structure, a variable length genotype may be able to learn not only the parameters of the solution, but also how many parameters there should be, what they mean, and how they interrelate

The variable length genotype is perhaps GP’s most radical practical innovation compared to its EA roots

Page 5: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

5

7.3 Iteration, Selection, and 7.3 Iteration, Selection, and Variable Length Program StructuresVariable Length Program Structures

The essence of evolution iterative insofar as generation after generation of population

s are assigned reproduction opportunities selective, insofar as the better performing variants get a bett

er chance to use these opportunities

Dawkins called this aspect cumulative selection the effects of selection acting on one generation are inherite

d by the next

Page 6: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

6

7.4 Evolvable Representations7.4 Evolvable Representations

Problem representation most ML paradigm: a fairly constrained problem

representation (Boolean, threshold, decision trees, etc)

Constraints advantage: making the traversal of the solution space more

tractable as long as the solution space is well tailored to the problem domain

GP search space the problem space + the space of the representation of the

problem

Page 7: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

7

7.4 Evolvable Representation (2)7.4 Evolvable Representation (2)

Ignoring Operators or Terminals system should magnify the exploration of sections of the rep

resentation space that produce better results

Finding Solutions of the Correct length GP can find a short or a long solution where a fixed length re

presentation cannot

Modularization and Meta-Learning 표현방법 탐색 공간에 대한 연구 Modularization (chap 10) Meta-Learning

information about the problem representation from one GP run is used to bias the search in later GP runs

Page 8: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

8

7.5 The Emergence of Introns, 7.5 The Emergence of Introns, Junk DNA, and BloatJunk DNA, and Bloat

Angeline the first GP researcher to associate this emergent “extra cod

e” in GP with the concept of biological introns “occasional occurrence”

Tackett GP bloat was caused by blocks of code in GP individuals

A body of research has established that GP bloat is, in reality, caused by GP introns introns are a persistent and problematic part of the GP proce

ss

Page 9: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

9

7.5.1 What Can We Learn 7.5.1 What Can We Learn from Biological Intronsfrom Biological Introns

생물학적 개념의 introns 과 GP 의 introns 은 거의 동일한 개념 phenotype 에 아무런 직접적 영향을 주지 않음 play some role in protecting good building blocks against de

structive crossover have indirect effect on survivability, the nature of the effect is

different

Page 10: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

10

7.6 Introns in GP Defined7.6 Introns in GP Defined

Introns in GP A feature of the genotype that emerges from the process of the ev

olution of variable length structures does not directly affect the survivability of the GP individual

properties introns are emergent and they do not directly affect the fitness of t

he individual

artificially inserted introns: artificial intron equivalents Global

introns for every valid input to the program

local only for the current fitness cases and not necessarily for other vali

d inputs

Page 11: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

11

7.7 Why GP Introns Emerge7.7 Why GP Introns Emerge

While introns do not affect the fitness of the individual, they do affect the likelihood that the individual’s descendents will survive

effective fitness survivability of an individual’s offspring The fitness of the parent. The fitter the parent, the more likel

y it is to be chosen for reproduction the likelihood that genetic operators will affect the fitness of t

he parent’s children

introns emerge principally in response to the frequently destructive effects of genetic operators

Page 12: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

12

7.8 Effective Fitness and Crossover7.8 Effective Fitness and Crossover

Complexity of the program length or size of the program measured with a method that

is natural for a particular representation

absolute complexity total size of the program or block

effective complexity the length of the active parts of the code within the program

Page 13: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

13

7.8 (2)7.8 (2)

Using fitness-proportion selection, block exchange crossover

Cje : the effective complexity of program j

Cja : the absolute complexity of program j

pc : probability of crossover

pjd : probability of destructive crossover

fj : the fitness of the individual

bar ft : average fitness of the population the proportion of copies of a program in the next generation is the

proportion produced by the selection operator minus the proportion of program destroyed by crossover

Page 14: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

14

7.8 (3)7.8 (3)

Rewrite

effective fitness

increase its effective fitness by lowering its effective complexity

Page 15: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

15

7.9 Effective Fitness and Other Operator7.9 Effective Fitness and Other Operator

Generalization to include the effects of other operator

Page 16: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

16

7.10 Why Introns Grow Exponentially7.10 Why Introns Grow Exponentially

Introns can provide very effective global protection against destructive crossover

최적해에 근접한 개체는 fitness 를 증가시키는 것은 매우 힘들어지고 , 최소한 현상태를 유지시키기 위해서 Effective fitness 를 증가시키는 방향을 선택한다 . Introns 증가 한계가 주어지지 않는다 .

Page 17: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

17

7.11 The Effects of Introns7.11 The Effects of Introns

introns may benefit evolution vs. introns almost always result in poor evolution and extended computation

issues about introns Introns may have differing effects before and after exponenti

al growth of introns begins Different systems may generate different types of introns wit

h different porbabilities The extent to which genetic operators are destructive in their

effect is likely to be a very important initial condition in intron growth

mutation and crossover may affect different types of introns differently

Page 18: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

18

7.11.1 Problems Caused by Introns7.11.1 Problems Caused by Introns

Run stagnation, poor results, and a heavy drain on memory and CPU time run stagnation: mutation 으로 해결

intron 을 의미있는 코드로 변화시킴 poor results: introns 의 확장을 제한함

Page 19: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

19

7.11.2 Possible Beneficial 7.11.2 Possible Beneficial Effects of IntronsEffects of Introns

To promote parsimony in the real code

a high probability of destructive crossover some introns in the population a system that makes it relatively easier to reduce the amoun

t of effective code than to add more introns

Structural protection against crossover

Page 20: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

20

7.12 What to Do about Introns7.12 What to Do about Introns

Reduction of destructive effects problem of bloat may be viewed in a more general way as

the absence of homology in GP

Parsimony the effect of parsimony pressure is to attach a penalty to the

length of programs see next slide

Changing the fitness function the fitness function become variable -> GP individuals might

find ways to improve their fitness

Page 21: 7. Genetic Programming and Emergent Order GP-Seminar 98.9.19 신수용

21

7.12 (2)7.12 (2)

The effect of parsimony