File Download

There are no files associated with this item.

  Links for fulltext
     (May Require Subscription)
Supplementary

Conference Paper: Gaussian mixture model of evolutionary algorithms

TitleGaussian mixture model of evolutionary algorithms
Authors
Issue Date2014
PublisherAssociation for Computing Machinery. The Proceedings' web site is located at http://dl.acm.org/citation.cfm?id=2576768
Citation
The 14th Annual Conference on Genetic and Evolutionary Computation Vancouver, BC, Canada, 12-16 July 2014. In GECCO'14: Proceedings of the 2014 Genetic and Evolutionary Computation Conference, p. 1423-1430 How to Cite?
AbstractThis paper proposes a novel finite Gaussian mixture model to study the population dynamics of evolutionary algorithms on continuous optimization problems. While previous research taking on a dynamical system view has established the transition equation between the density functions of consecutive populations, the equation usually does not have closed-form solutions and can only be applied to very few optimization problems. In this paper, we address this issue by approximating both the population density function of each generation and the objective function by finite Gaussian mixtures. We show that by making such approximations the transition equation can be solved exactly and key statistics, such as the expected mean and the variance of fitness values of the population, can be calculated easily. We also prove that by choosing appropriate values of the parameters, the $L^1$-norm error between our model and the actual population density function can be made arbitrarily small, up until a predefined generation. We present experimental results to show that our model is useful in simulating and examining the dynamics of evolutionary algorithms.
Persistent Identifierhttp://hdl.handle.net/10722/217397
ISBN

 

DC FieldValueLanguage
dc.contributor.authorSong, B-
dc.contributor.authorLi, VOK-
dc.date.accessioned2015-09-18T05:58:23Z-
dc.date.available2015-09-18T05:58:23Z-
dc.date.issued2014-
dc.identifier.citationThe 14th Annual Conference on Genetic and Evolutionary Computation Vancouver, BC, Canada, 12-16 July 2014. In GECCO'14: Proceedings of the 2014 Genetic and Evolutionary Computation Conference, p. 1423-1430-
dc.identifier.isbn978-1-4503-2662-9-
dc.identifier.urihttp://hdl.handle.net/10722/217397-
dc.description.abstractThis paper proposes a novel finite Gaussian mixture model to study the population dynamics of evolutionary algorithms on continuous optimization problems. While previous research taking on a dynamical system view has established the transition equation between the density functions of consecutive populations, the equation usually does not have closed-form solutions and can only be applied to very few optimization problems. In this paper, we address this issue by approximating both the population density function of each generation and the objective function by finite Gaussian mixtures. We show that by making such approximations the transition equation can be solved exactly and key statistics, such as the expected mean and the variance of fitness values of the population, can be calculated easily. We also prove that by choosing appropriate values of the parameters, the $L^1$-norm error between our model and the actual population density function can be made arbitrarily small, up until a predefined generation. We present experimental results to show that our model is useful in simulating and examining the dynamics of evolutionary algorithms.-
dc.languageeng-
dc.publisherAssociation for Computing Machinery. The Proceedings' web site is located at http://dl.acm.org/citation.cfm?id=2576768-
dc.relation.ispartofGECCO'14: Proceedings of the 2014 Genetic and Evolutionary Computation Conference-
dc.rightsGECCO'14: Proceedings of the 2014 Genetic and Evolutionary Computation Conference. Copyright © Association for Computing Machinery.-
dc.titleGaussian mixture model of evolutionary algorithms-
dc.typeConference_Paper-
dc.identifier.emailSong, B: bosong@HKUCC-COM.hku.hk-
dc.identifier.emailLi, VOK: vli@eee.hku.hk-
dc.identifier.authorityLi, VOK=rp00150-
dc.identifier.doi10.1145/2576768.2598252-
dc.identifier.hkuros254361-
dc.identifier.spage1423-
dc.identifier.epage1430-
dc.publisher.placeNew York, NY-

Export via OAI-PMH Interface in XML Formats


OR


Export to Other Non-XML Formats