二进神经网络学习算法研究

把二进神经网络对布尔函数映射的学习归结为神经元对学习样本集合的表达。通过对神经元表达能力的分析研究，引入加权距离汉明球的概念，既提高了学习效率也简化了布尔函数实现结构。同时把汉明球及立方体集合覆盖思想等统一在加权汉明距离球覆盖的框架下，另外，还得到旨在提高输出层神经元表达能力的新结果，最后举例说明了此学习策略的可行性与特点，经学习得到的二进神经网络的权系数及阈值皆为整数，易于硬件实现。

RESEARCH ON THE LEARNING ALGORITHM OF BINARY NEURAL NETWORK

In this paper, the learning of binary neural network for Boolean function mapping is projected as representing sets of the learning patterns by neurons. For hidden layer, a concept of weighted Hamming distance sphere is introduced to further improve the learning efficiency and generalize the learning strategies as well. For output layer, some new results about representing combination of sets, which are devoted to simplify implementation of Boolean function, are proposed. Under the learning strategies the weights and thresholds are all integers, and the network is easy to be implemented by hardware.