摘要
本文提出了神经网络(NN)知识表示的形式化描述语言和知识单元的概念,用于给出传统符号逻辑中的概念、属性及它们之间的层次关系如何在神经网络中进行表示,提出了激活强度的度量标准,从而在理论上给出了NN中继承和识别问题的形式化处理方法,在此基础上,提出了NN正向和反向问题求解机制,这里,值得一提的是,本文通过在某一特定领域中对NN推理机制的成功探讨,有力地反驳了人们对于NN模型的一个公开批评,即:它不能表达高度结构化的知识并在这种知识上进行有效的推理。
The paper proposes a formal description language for neural networks knowledge representation and the concept of knowledge unit,to represent the concepts,attributes,and their layered relationship.The measuring standard of stimulus strength,the formal method of inheritance and recognition in neural networks,and the neural networks positive and negative problem solving mechanism are presented.This work effectively refutes the common criticism that neural networks can not represent highly structured knowledge and infer on these knowledge.
出处
《计算机学报》
EI
CSCD
北大核心
1993年第11期814-822,共9页
Chinese Journal of Computers
关键词
知识单元
激活强度
神经网络
Knowledge unit,stimulus strength,inheritance and recognition,inference,neural networks.