进化神经网络的编码表示机制研究

A Study on Encoding Representations of Evolving Neural Networks

神经网络学习包括权值学习和拓朴结构学习,目前大部分妍究集巾于权值学习,结构的选择和确定一般是靠经验和专家知识。设计一个好的神经网络招扑结构是非常困难的。一般地,学习,连同选择一个优化网络招扑结构,是一个NP宇全问题。

This paper discusses the classification and properties of the encoding representations of e-volving neural networks,analyses the advantages and disadvantages of these methods. These would help designer select encoding representations in different applications.