神经网络的规则提取研究

AN APPROACH TO RULE EXTRACTION OF NEURAL NETWORKS

文中论述了作为解决神经网络“黑箱问题”有效手段的规则提取方法，分析了基于结构分解和输入输出映射的神经网络规则提取的各种算法，概括了它们的基本思想并分析了它们的优劣，在相似权值法的基础上提出 Ｃ Ｓ Ｗ 算法，有效解决了连续值输入网络的规则提取问题．将 Ｃ Ｓ Ｗ 算法应用于 Ｉ Ｒ Ｉ

In this paper, the rule extraction of neural networks is disscussed, which is an effective method to avoid the shortcoming of being “black boxes”. Techniques based on decompositional and input output mapping approaches are studied and their fundamental concepts and evaluates their performances are generalized. Based on similar weight approach, the CSW approach is proposed to efficiently solve the rule extraction from continuous\|input neural networks. CSW is applied in IRIS Flower Classification Problem,and experiment results show that rules extracted by our method are accurate and comprehensible.