摘要
本文以自适应步长调整代替ＤＦＰ优化方法中的一维搜索，并将它应用于前馈神经网络训练中。同时对步长的衰减因子和增长因子进行动态自适应调整，提高了ＢＰ训练算法的收敛速度和摆脱局部极小的能力，实验结果证明了该算法的有效性。
In this paper,the DFP optimal method is improved by replacing the 1D seek with the learning rate adaptive adjusting,and the method is applied to training FNN.In addition,both the increment factor and the decrement factor are adjusted adaptively and dynamically.Thus,both the convergent rate and the ability to escape from local minimum.Experimental results show the effectiveness of the proposed training algorithm.
出处
《计算机工程与科学》
CSCD
1998年第4期17-20,共4页
Computer Engineering & Science
