二进神经网络的模式匹配学习

二进神经网络的知识提取需要了解每个神经元的逻辑意义。一般来说,对二进神经网络学习结果的分析是困难的。该文提出了一种基于线性可分结构系结构分析的学习算法,采用这种方法对布尔空间的样本集合进行学习,得到的二进神经网络隐层神经元都归属于一类或几类线性可分结构系,只要这几类线性可分结构系的逻辑意义是清晰的,就可以分析整个学习结果的知识内涵。

The pattern match learning of binary neural networks

It is necessary to know the logical meaning of every binary neuron when extracting knowledge from a binary neural network. Generally, it is difficult to analyze learning results of a learning algorithm for binary neural networks. In this paper, a new learning method is presented which is based on analyzing a set of linear separable structures. The most important benefit of this method is all binary neurons belong to one or more types of linear separable structure sets. If those linear separable structure sets have clear logical meaning, the whole knowledge of binary neural networks can be dug out.