1 前言各种神经网络被广泛用来解决有导师分类问题.对于模糊感知器,文献[1]提出一种新的训练算法,并证明当样本可分时,该算法有限收敛.文献[2]在模糊神经元中加入阈值 v∈[0,1].对某些模糊神经网络学习算法如FBP(见文献[3]),阈值在收敛性证明中起着重要的作用.那么,在模糊感知器中加入阈值是否还能得到算法的收敛性?本文将对这个问题进行讨论.
A modified algorithm for a fuzzy perceptron with bias is presented, and its finite convergence is proved in the case that the fuzzy training patterns are separable. An illustrative numerical result is included and is compared with the method without bias.