BP神经网络收敛性问题的改进措施

The convergence and improvements of BP neural network

BP算法现在已成为目前应用最广泛的神经网络学习算法,它在函数逼近、模式识别、分类、数据压缩等领域有着更加广泛的应用,但存在收敛较慢问题.笔者在文中简述了BP算法原理,针对BP算法的收敛性问题,提出了几点改进措施.

The BP(back propagation) algorithm is a neural network learning algorithm, it is applied extensively in function approximation, mode distinguishing, classification, data compression et, but it has a question of convergence. In this paper, based on describing the principle of the BP algorithm, the convergence is discussed deeply, and several improvements to BP neural network are proposed.