奇偶校验问题的二进神经网络学习算法

二进神经网络可以完备表达任意布尔函数,但对于孤立节点较多的奇偶校验问题却难以用简洁的网络结构实现。针对该问题,提出了一种实现奇偶校验等孤立节点较多的一类布尔函数的二进神经网络学习算法。该算法首先借助蚁群算法优化选择真节点及伪节点的访问顺序;其次结合几何学习算法,根据优化的节点访问顺序给出扩张分类超平面的步骤,从而减少隐层神经元的数目,同时给出了隐层神经元及输出元的表达形式;最后通过典型实例验证了该算法的有效性。

Learning Algorithm of Binary Neural Networks for Parity Problems

Binary neural network can completely express arbitrary Boolean function, but more isolated nodes such asparity problem are difficult to implement with simple network structure. According to this problem, we presented alearning algorithm to realize 13oolcan function such as parity problems with many isolated samples. 13y means of the antcolony algorithm, we obtained the optimized core nodes and the extension order of true and false nodes, by combing thegeometrical algorithm, we gave the steps of how to expand the classifier hyperplanes with the optimized core nodes, sothis algorithm can reduce the number of hidden neurons in network, and the expression of the hidden neurons and theoutput neuron are also given. Finally,this algorithm is validated to be effective through examples.