论文部分内容阅读
本文首先给出二进前向多层网几何学习算法[1,2]的一个改进策略,提高了原算法的学习效率.然后提出一个新的神经网络启发式遗传几何学习算法(简称HGGL算法).H~算法采用面向知识的交叉算子和变异算子对几何超平面进行优化的划分,同时确定隐层神经元的个数及连接权系数和阈值对任意布尔函数,HGGL算法可获得迄今为止隐节点数最少的神经网络结构
In this paper, we give an improved strategy of the binary learning geometric learning algorithm [1, 2], and improve the learning efficiency of the original algorithm. Then a new neural network heuristic genetic geometry learning algorithm (HGGL algorithm) is proposed. H ~ algorithm using knowledge-oriented crossover operator and mutation operator to optimize the geometric hyperplane division, while determining the number of hidden neurons and connection weights and thresholds For any Boolean function, HGGL algorithm can be obtained so far Neural network structure with the least number of nodes