Author:
Jeffrey Horn
Affiliation:
Northern Michigan University, United States
Keyword(s):
Shape nesting, Coevolution, Co-evolution, Cooperation, Cooperative Co-evolution, Fitness sharing, Resource-defined fitness sharing, Speciation, Niching, Selection, Infinite population.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Bio-inspired Hardware and Networks
;
Co-Evolution and Collective Behavior
;
Computational Intelligence
;
Concurrent Co-Operation
;
Evolutionary Computing
;
Genetic Algorithms
;
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Soft Computing
;
Swarm/Collective Intelligence
Abstract:
We test the effectiveness of an evolutionary algorithm that relies completely on species selection for evolution and on interactions among species to determine fitness. Under Resource-defined Fitness Sharing (RFS), all individuals have the same objective fitness, but they act to reduce their shared fitness through competition for resources. In previous studies, RFS has been used to evolve populations of mutually non-competing (i.e., non-overlapping) shapes on shape nesting problems. In this paper we test the effectiveness of a modified version of RFS, which we call PCSN, against three commercial software packages for shape nesting. PCSN uses species proportions to represent a population, thereby simulating an infinitely large population. With no discovery operators, such as mutation or recombination, evolution consists of selection only, with all species present in the initial population. We show that on some shape nesting problems this approach can outperform some commercial
packages. In particular, PCSN nests more circles on a fixed, polygonal substrate than do most of the commercial packages. This might be considered a surprising result, since the algorithm is radically different from any shape nesting algorithms deployed to date. While conventional methods place one shape at a time, the co-evolution approach attempts to place all shapes simultaneously.
(More)