摘要
提出一种新的基于神经网络的规则提取方法 .提出的网络由一个主网络及其映射网络组成 ,具有二次收敛过程 .通过主网络的学习 (第 1次收敛 )完成知识学习和网络构造 ,在此基础上构造了其网络映射 ,通过该映射网络的收敛过程实现规则的提取 .该方法在规则提取时无须遍历解空间 ,从而很好地提高了搜索效率 ,降低了计算复杂度 .同时 ,还提出估计规则数下限的信度差方法 .
A novel neural network based rule extraction method is proposed in this paper. This method consists of a primary network and its corresponding mapping network, which includes twice convergent processes. The knowledge acquisition and network construction of the method are fulfilled by the first convergence of the primary network. Here by a mapping network corresponding to the converged primary network is created whose convergence is capable of realizing the rule extraction. Since there is no need of enumerating the overall space of solutions for this method to extract rules, therefore the searching efficiency is greatly increased and the computation complexity is dramatically reduced. Meanwhile, a stop criterion of rule extraction in terms of difference of belief degree is also proposed in this paper. A lot of simulation experiments and practical applications illustrate and verify the validity and correctness of the proposed method.
出处
《软件学报》
EI
CSCD
北大核心
2000年第12期1635-1641,共7页
Journal of Software
基金
国家自然科学基金资助项目(69835001)