cooperative coevolution : an architecture for evolving coadapted subcomponents

12
MITCHELL A. POTTER KENNETH A. DE JONG SPEAKER: LIN, WEI-KAI (2008.12.16) Cooperative Coevolution: An Architecture for Evolving Coadapted Subcomponents

Upload: marlin

Post on 24-Feb-2016

40 views

Category:

Documents


0 download

DESCRIPTION

Cooperative Coevolution : An Architecture for Evolving Coadapted Subcomponents. Mitchell A. Potter Kenneth A. De Jong Speaker: Lin, Wei-Kai (2008.12.16). Introduction. To appear in Evolutionary Computation 8(1):1-29, 2000, The MIT Press - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

MITCHELL A . POTTERKENNETH A. DE JONG

SPEAKER: L IN, WEI-KAI (2008.12 .16)

Cooperative Coevolution: An Architecture for Evolving Coadapted Subcomponents

Page 2: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Introduction

To appear in Evolutionary Computation 8(1):1-29, 2000, The MIT Press

After the cooperative coevolutionary framework published in 1994

One of the major difficulties comes in finding subcomponents to “emerge” rather than being hand designed Evolving coadapted subcomponents

Page 3: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Cooperative Coevolution Architecture

Page 4: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Problem Decomposition

Both the number of species and in the ecosystem and the roles the species assume to be an emergent property of cooperative coevolution

“If evolution stagnates, it may be that there are too few species in the ecosystem from which to construct a good solution”

Initialize the species randomly and evaluate its individuals based only on the overall fitness of the ecosystem

Page 5: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

The Algorithm

A species finds a niche where it can make a contribution, it will tend to exploit this area

If a species is unproductive, determined by the contribution its individuals make to the collaborations they participate in, the species will be destroyed

Otherwise it is keptStagnation can be detected

GLtftf )()(

Page 6: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Coevolving String Covers

Matching set M (of size N) and target set of size K

Different targets

Page 7: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Results

Given the number of species

Page 8: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Case Study in Emergent Problem Decomposition

Evolving cascade networks

Page 9: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Two-Spirals Problem

Black points (training data)

Page 10: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Experimental Results

Page 11: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Comparison

Page 12: Cooperative  Coevolution : An Architecture  for Evolving  Coadapted  Subcomponents

Conclusions and FutureWork

An architecture for evolving subcomponents as a collection of cooperating species

Explored the emergence of coadapted subcomponents in more detail through a case study involving neural network evolution