AB network adjust the step and the hidden-layer neurons algorithm based on BP network
For the classical BP algorithm has some deficiencies, such as the accuracy is insufficient, the rate of convergence does not descend, weight value closes to zero. This paper proposes the AB neural network to adjust the step and the hidden-layer neurons algorithm based on BP network. Network A with learning ability, configures and adjusts the structure of Network B and trains it, by adjusting the step and the hidden-layer neurons of Network B, obviously enlarge the modification of weight to escape from flat region. The introduction of prior knowledge made training of Network B intelligently and automatically. The simulation results of Sin Function shows that the proposed method can effectively speed up the multilayer feed-forward neural network training process.
AB neural network step hidden-layer neurons prior knowledge
Ningsheng Gong Yan Liu
College of Information Science and Engineering Nanjing University of Technology Nanjing, Jiangsu,Chi College of Information Science and Engineering Nanjing University of Technology Nanjing, Jiangsu, Ch
国际会议
重庆
英文
536-539
2011-08-20（万方平台首次上网日期，不代表论文的发表时间）