UTokyo Repository 東京大学

UTokyo Repository >
113 工学系研究科・工学部 >
22 電気系工学専攻 >
1132220 博士論文(電気系工学専攻) >

このページ(論文)をリンクする場合は次のURLを使用してください: http://hdl.handle.net/2261/52452

タイトル: Learning Potential Inheritance in Baldwinian Evolution
その他のタイトル: Baldwin進化における学習能力の遺伝
著者: Liu, Shu
著者(別言語): リュウ, ジュ
発行日: 2012年9月27日
抄録: Inspired by organism evolution, evolutionary algorithms have attracted research interests for several decades, and have been proved effective and efficient to solve optimization problems. For further enhancement, local search are introduced into evolution, imitating the biological fact that organism individuals change themselves to better fit the environment during their lifetime. In the context of evolutionary computation, Baldwinian evolution hybridizes global search through population-based evolution with local search through individual refinements. Refinement influences selection, however, unlike Lamarckian evolution, refined traits are not inherited by the offspring. In Baldwinian evolution, local search guides evolution in an indirect manner, termed "the Baldwin effect". Conventional studies revealed that Baldwinian learning can enhance search, however, there is still considerable work to be done to understand the mechanisms. This thesis investigates Baldwinian evolution's mechanisms in depth. By proposing the method of analyzing individual dynamics, we present some novel views. We find that the substantial inheritance in Baldwinian evolution is the potential to achieve high fitness through learning, that the realization of learning potential is influenced by inheritable and noninheritable factors in evolution, and that learning cost penalties inhibit learning intensity. Our results provide knowledge of Baldwinian evolution's mechanisms, and directions to possible applications.
内容記述: 報告番号: ; 学位授与年月日: 2012-09-27 ; 学位の種別: 課程博士 ; 学位の種類: 博士(工学) ; 学位記番号: ; 研究科・専攻: 工学系研究科電気系工学専攻
URI: http://hdl.handle.net/2261/52452
出現カテゴリ:021 博士論文
1132220 博士論文(電気系工学専攻)


ファイル 記述 サイズフォーマット
37097408.pdf1.89 MBAdobe PDF見る/開く



Valid XHTML 1.0! DSpace Software Copyright © 2002-2010  Duraspace - ご意見をお寄せください