用学习和搜索方法从神经元网络中抽取知识

Using two phase approach to extract knowledge from artificial neural networks

针对训练好的神经元网络进行解释这一难以解决的问题，提出了一种从神经元网络中抽取规则的新的抽取方法——二阶段法，从隐含层到输出层，利用学习方法从整个隐含激励空间中抽取出有效区域，形成规则；从输入层到隐含层，利用搜索方法，通过分析其间的权值关系抽取出规则，使得所有被这些规则覆盖的实例所产生的隐含激励向量均位于上述有效区域内。实验证明了此方法产生的规则比Ｃ４．５产生的规则的抗干扰力强，同时其可信度比传统的基于搜索的抽取方法——ＫＴ算法要高。

According to the understanding problem of artificial neural networks, this paper prvide a new method of extracting rules from trained heural network——two phases algorithm. From hidden layer to output layer, it extracts rules by identifying valid regions in the whole hidden activation space. From input layer to hidden layer, it searches the rules based on the analysis of weights between input nodes and hidden nodes so that all instances covered by these rules generate hidden activation vectors lying in the valid regions. Some experiments have demonstrated that our method generates rules of better performance than the decision tree approach in noisy conditions and the fidelity of the rules extracted from a neural network with distributed representations in our method is higher than that in conventional search based methods, such as KT algorithms.