基于蚁群算法的神经网络规则抽取  

Rules Extraction from Trained Neural Networks Based on Ant Colony Algorithm

作　　者：高在村[1,2] 邓伟[1] 

机构地区：[1]苏州大学计算机与科学技术学院,苏州215006 [2]扬州职业大学信息工程学院,扬州225000

出　　处：《微电子学与计算机》2009年第5期156-159,162,共5页Microelectronics & Computer

基　　金：国家自然科学基金项目(60673092);教育部科研重点项目(205059);江苏省高校自然科学基金(07KJD520186)

摘　　要：从神经网络的功能性观点出发,将蚁群算法用于神经网络的规则抽取,为每个隐单元和输出单元生成各自的规则,然后依照网络的传导方向聚合这些规则,从而为整个网络抽取出理解性好、简洁的符号规则.该方法不依赖于具体的网络结构和训练算法,可以方便地应用于各种分类器型神经网络.实验结果表明了该方法的实用性和可行性.This paper from the functional point of view,proposed a method that the ant colony algorithm is applied to the rule extraction from neural networks,generating respective rules for each hidden units and output units then polymerizing these rules according to the network＇s conduction direction,thus extracting accurate,concise and comprehensible symbolic rules for the whole network.This method is independent of the architecture and training algorithm so that it could be easily applied to diversified neural classifiers.The result of experiment has shown its practicability and feasibility.

关 键 词：神经网络 规则抽取 蚁群算法 聚类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]