摘要
本文讨论了一种用于模式分类的高阶神经网的快速学习法.该方法分为二个阶段,首先是样本空间的划分,然后由样本及它所对应的类训练权值.学习过程没有迭代循环,一次确定.隐元的个数有明确含义,由样本空间决定,而不是由人判断.该学习方法十分简单,适合于硬件实现.本文中给出了该方法应用于XOR问题,对称性判别,T—C分类问题的例子.以及该模型的逻辑电路实现方法.
A fast learning algorithm for multi-layer neural nets is presented. The learning is with two steps. First, dividing the sample space by the front layer weights; Second,training the latter layer weights by input samples. The learning process is non cycle, and non iterating, and the number of the hidden units is determined by the sample space not by persons.The model is quite adapted to hardware realization. Some examples and logical circuit design are given in the paper.
出处
《信号处理》
CSCD
北大核心
1991年第3期129-134,共6页
Journal of Signal Processing
