摘要
针对神经网络的发展状况,利用概率及其它数学的方法,对神经网络中BP算法进行了分析,对BP算法的学习速度慢、容错能力差、算法不完备等缺点进行了讨论,得出BP神经网络的这些局限性均是BP算法本身固有的。在此基础上提出新的算法TFP,并讨论了网络假吸引中心的问题,在网络性能上,对保持知识的稳定性也作了探讨。
Using probability and other mathematical method, the BP neural network algorithm was analyzed. The defects such as slow learning, worse fault-tolerant capability and nonperfect of algorithm BP algorithm were discussed. These limitations are inherent of BP algorithm itself. On the basis of the discussion, new TFP algorithm was suggested. The problerms of spurious attractive centre in the network and maintinance of the knowledge stability in the network performance were also discussed.
出处
《山西农业大学学报（自然科学版）》
CAS
2009年第1期89-93,共5页
Journal of Shanxi Agricultural University(Natural Science Edition)
基金
山西省自然科学基金(20041010)
关键词
BP网络
局部极小点
梯度下降法
BP Neural Network
Partial minimum point
Gradient drop law
