用最优化方法解决BP神经网络训练问题

BP神经网络可以有效地对非线性系统进行逼近,但是传统的最速下降搜索方法存在收敛速度慢的问题。本文提出把BP神经网络转化为最优化问题,用一种共轭梯度算法代替最速下降法进行搜索迭代,极大地提高了收敛速度。

Training BP Neural Network Using Optimization Methods

BP neural network can efficiently approximate any nonlinear system, but there is a problem of inefficient learning speed with the conventional steepest descent algorithm. In this paper, we try to convert neural network to an optimization model, and apply conjugate gradient algorithm to it to bring a faster learning speed.