| BP自适应学习率设计 |
| |
| 引用本文： | 杨启文,韩玉兵.BP自适应学习率设计[J].河海大学常州分校学报,2001,15(3):20-24. |
| |
| 作者姓名： | 杨启文 韩玉兵 |
| |
| 作者单位： | 1. 河海大学,计算机及信息工程学院,江苏,常州,213022
2. 河海大学常州校区,数理教学部,江苏,常州,213022 |
| |
| 摘 要： | 在神经网络的训练当中存在“过学习”现象以及参数难以选择的困难 .本文指出了大多数自适应 BP算法在学习率自调整中存在的问题 :事后矫正 ,同时通过极其简单的优化手段 ,为当前权值的每一次调整选择一个次优 (甚至最优 )的学习率 ,从而改善了训练过程的平稳性和快速性 ,并且有效地解决了神经网络的“过学习”问题
|
| 关 键 词： | 神经网络 自适应学习率 过学习 反传算法 |
| 文章编号： | 1009-1130(2001)03-0020-05 |
| 修稿时间： | 2000年11月14 |
Design of Adaptive Learning Rate for BP Algorithm |
| |
| Yang Qi wen ,Han Yu bin.Design of Adaptive Learning Rate for BP Algorithm[J].Journal of Hohai University Changzhou,2001,15(3):20-24. |
| |
| Authors: | Yang Qi wen Han Yu bin |
| |
| Institution: | Yang Qi wen 1,Han Yu bin 2 |
| |
| Abstract: | There exist over learning and the difficulties of selecting suitable parameters when training neural network.In this paper,the deficiency of adaptive BP algorithms in regulating learning rate is pointed out as the so called optimal parameters for the current iteration aren't updated until the next iteration.We choose a sub optimal(even an optimal)learning rate for the current regulation of weights in every iteration by means of a highly simple method.As a result,the stability and speediness of the training process are improved and the problem of over learning is also solved effectively. |
| |
| Keywords: | neural network adaptive learning rate over learning back propagation algorithm |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
