论文部分内容阅读
该文提出了一种基于超立方体覆盖的构造性神经网络学习算法,以解决二值型输入变量的K分类问题。该算法分两步来动态地构造一个三层前馈网络。首先,对于每一类的所有训练样本,用尽可能少的超立方体来覆盖它们,并为每一个超立方体构造一个隐层单元;其次,用“或”操作把这些隐单元连接到相应的输出单元上。文章给出了相应的理论分析和一个具体的实现。实验结果表明,该算法优于常用的一些归纳学习算法。
In this paper, a learning algorithm based on hypercube covering constructive neural network is proposed to solve the K classification problem of binary input variables. The algorithm dynamically builds a three-layer feedforward network in two steps. First, for all training samples in each class, cover them with as few hypercubes as possible and construct a hidden layer cell for each hypercube. Second, connect these hidden cells to the corresponding On the output unit. The article gives the corresponding theoretical analysis and a concrete realization. Experimental results show that this algorithm is superior to some commonly used inductive learning algorithms.