文章基本信息

标题：ALENA: Adaptive-Length Evolving Neural Arrays
本地全文：下载
作者：L. C. Corbalán ; L. Lanzarini ; A. E. De Giusti 等
期刊名称：Journal of Computer Science and Technology
印刷版ISSN：1666-6046
电子版ISSN：1666-6038
出版年度：2004
卷号：4
期号：1
出版社：Iberoamerican Science & Technology Education Consortium
摘要：Evolving neural arrays (ENA) have proved to be capableof learning complex behaviors, i.e., problems whosesolution requires strategy learning. For this reason, theypresent many applications in various areas such as roboticsand process control. Unlike conventional methods –basedon a single neural network– ENAs are made up of a set ofnetworks organized as an array. Each of them represents apart of the expected solution.This work describes a new method, ALENA, thatenhances the solutions obtained by solving the maindeficiencies of ENA since it eases the obtaining ofspecialized components, does not require the explicitdecomposition of the problem into subtasks, and iscapable of automatically adjusting the arrays length foreach particular use.The measurements of the proposed method –applied toproblems of obstacle evasion and objects collection– showthe superiority of ALENA in relation to the traditionalmethods that deal with populations of neural networks.SANE has been used in particular as a comparativereferent due to its high performance. Eventually,conclusions and some future lines of work are presented
关键词：Evolving Neural Networks; Evolving Neural;Arrays; Learning; Genetic Algorithms; Subpopulations