Display Publications by [Year] [Type] [Topic]
Evolutionary methods are powerful tools in discovering solutions for difficult continuous tasks. When such a solution is encoded over multiple genes, a genetic algorithm faces the difficult credit assignment problem of evaluating how a single gene in a chromosome contributes to the full solution. Typically a single evaluation function is used for the entire chromosome, implicitly giving each gene in the chromosome the same evaluation. This method is inefficient because a gene will get credit for the contribution of all the other genes as well. Accurately measuring the fitness of individual genes in such a large search space requires many trials. This paper instead proposes turning this single complex search problem into a multi-agent search problem, where each agent has the simpler task of discovering a suitable gene. Gene-specific evaluation functions can then be created that have better theoretical properties than a single evaluation function over all genes. This method is tested in the difficult double-pole balancing problem, showing that agents using gene-specific evaluation functions can create a successful control policy in 20\% fewer trials than the best existing genetic algorithms. The method is extended to more distributed problems, achieving 95\% performance gains over tradition methods in the multi-rover domain.
@inproceedings{tumer-agogino-miik_gecco05,
title = {Efficient Credit Assignment through Evaluation Function Decomposition},
author = {A. Agogino and K. Tumer and R. Miikulainen},
booktitle = {The Genetic and Evolutionary Computation Conference},
month = {June},
address = {Washington, DC},
abstract = {Evolutionary methods are powerful tools in discovering solutions for difficult continuous tasks. When such a solution is encoded over multiple genes, a genetic algorithm faces the difficult credit assignment problem of evaluating how a single gene in a chromosome contributes to the full solution. Typically a single evaluation function is used for the entire chromosome, implicitly giving each gene in the chromosome the same evaluation. This method is inefficient because a gene will get credit for the contribution of all the other genes as well. Accurately measuring the fitness of individual genes in such a large search space requires many trials. This paper instead proposes turning this single complex search problem into a multi-agent search problem, where each agent has the simpler task of discovering a suitable gene. Gene-specific evaluation functions can then be created that have better theoretical properties than a single evaluation function over all genes. This method is tested in the difficult double-pole balancing problem, showing that agents using gene-specific evaluation functions can create a successful control policy in 20\% fewer trials than the best existing genetic algorithms. The method is extended to more distributed problems, achieving 95\% performance gains over tradition methods in the multi-rover domain.},
bib2html_pubtype = {Refereed Conference Papers},
bib2html_rescat = {Evolutionary Algorithms, Optimization},
year = {2005}
}
Generated by bib2html.pl (written by Patrick Riley ) on Wed Apr 01, 2020 17:39:43