基于图论方法的前向神经网络优化结构的机理研究

we performed the project presented by the proposal very seriously, and made some extension about our research work beyond that proposal. In the research of this project, not only the relation between the neural networks and the graph theory was systematically studied, the stability of a neural network and the learning algorithm of a neural network were also dealt with, together with the study of the latestly developed DNA computation and DNA computer model, as well as genetic algorithm and its applications to the design of neural networks. The research results include more than 40 journal papers which were published respectively in Science in China Press, IEEE Trans. on Automatic Control, Discrete Mathematics, IEE Electronics Letters, Communications in Theoretical Physics, Natural Science Improvement, Computer Journal, Electronics Journal of Electronics (Chinese version and English version), Chinese Physics, Journal of Electronics and Information Technology, Applied Mathematics and Mechamics(English version), Applied Mathematics Journal(English version) and so fourth.

本项目拟以图论为工具研究前向神经网络结构优化中的核心问题-Boole 函数的线性与非线钥煞中晕侍?如线性可分Boole函数的计数与构造问题;复杂度>1的Boole函数的基本特征的袒⒓剖肮乖?其基本方法是先将Boole函数转化为对应的子图,然后通过刻划子图的特征来研究对应函数的分类复杂度的特征.最后应用所获结果于前向神经网络的结构优化.