摘要
本文对前馈神经网络的BP算法进行了分析,指出:目前使用的BP算法的主要缺点(学习速度慢、所得到的网络性能差、算法不完备)是BP算法本身固有的缺点,换句话说,不对BP算法进行本质的改变,是不可能克服上述缺点的。最后简单讨论一下,克服上述缺点的可能途径。
The performance of BP algorithm of feedforward neural networks have been discussed. We show that (1) the learning speed is quite low, (2) the performance performance of network constructed by the BP algorithm is poor, and (3) the algorithm is incomplete. All of the drawbacks above are inherent properties of BP algorithm.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1994年第3期191-195,共5页
Pattern Recognition and Artificial Intelligence
关键词
神经网络
BP算法
反馈网络
Muitilayered Feedforward Neural Network, BP Algorithm.
