登入選單

返回Google圖書搜尋

Designing Neural Networks

University of Illinois at Urbana-Champaign. Center for Supercomputing Research and DevelopmentGeorge Cybenko

出版	University of Illinois at Urbana-Champaign, Center for Supercomputing Research and Development, 1989
URL	http://books.google.com.hk/books?id=gKjPGwAACAAJ&hl=&source=gbs_api

註釋

Abstract: "In order to design and select neural network architectures appropriate for given recognition or classification tasks, we argue that it is necessary to develop a complexity theory for the underlying problems. A notion of complexity is proposed and some of its properties are developed. This notion of complexity determines an upper bound on the number of sigmoidal neural units needed at the first layer of a multilayer feedforward network. We also present an asymptotic upper bound on the complexity of classification problems where smooth boundaries separate the classes. A discussion of the application of these ideas to practical problems determined by empirical data sets is discussed."