零代价函数的前馈神经网络新算法

文章提出一种全新的三层前馈神经网络算法,它比BP算法优越得多.新算法不再采用误差反向传播原理,而是选择一层自由权,通过求解线性方程组求得另一层待求权.新算法选择很广一类的隐层神经元函数,可以直接求得全局最小点,不存在BP算法的局部极小、收敛速度慢等问题.新算法较BP算法快得多.

A Novel Algorithm for Feedforward Neural Networks with Zero Cost Function

： In this paper,a novel algorithm for three－layered feedforward neural networks is proposed,which is much better than BP algorithm.After the free weights are preassigned,the new algorithm can find the target weights of the other layer by solving linear equation sets.The algorithm can get global minimum easily with a wide variety of functions of hidden neurons,and no problems such as local minima and slow rate of convergence are suffered like BP algorithm.The new algorithm is much faster than BP algorithm．