摘要
作为将基于符号机制的逻辑推理与基于连接机制的神经网络的集成，逻辑神经网络研究具有重要的意义。一个逻辑函数唯一地确定了一个ｎ－维超立方体顶点的二分类，若对ｎ维超立方体的顶点进行正确分类，同时保证网络具有最好的容错能力，则分类超平面应过任两个不同类顶点连线的中点，基于这样的思想，本文导出了使网络容错能力最强的分类超平面的标准方程，给出了网络各层节点之间连接权值和偏置值的可能值，使得网络易于训练和实现，网络所去示的知识易于解释和易于进行规则提取。
As an integration of inference method and neural network method , logic neural network plays an important role in AI. In order to realize the correct classification of the apexes of n-dimensional hyperplane with a neural network of the maximum tolerance ability, the median point on the line of any two adjacent apexes falling into different classes must be in the hyperplane classifying these two apexes. Based on s,the standard equation of the hyperplane is induced, and the possible values of interconnective weights and biases are given, which result in maximum tolerant ability,much. easier to train and realize,and much easier to express and extract knowledge.
出处
《计算机仿真》
CSCD
1998年第3期24-26,64,共4页
Computer Simulation
基金
中国电子科学研究院基金
关键词
逻辑神经网络
符号机制
人工智能
遗传算法
Logic neural network n-Dimensional hypercube Classifying hyperplane
