适合于BP学习的共轭梯度法

CONJUGATE GRADIENT ALGORITHM SUITABLE FOR BP LEARNING

基于ＬＭＳ的标准ＢＰ算法收敛速度极慢，而共轭梯度法要求精确的线性搜索，这在神经网络的高维权空间中是难以实现的。本文提出了一种新的ＢＰ学习算法，它采用一种对线性搜索要求不高的改进的共轭梯度法与一种简单的不精确线性搜索相结合，极大地提高了ＢＰ学习速度。经多次测试表明，与标准ＢＰ算法相比，该算法的效率提高了二个数量极。

The LMS- based conventional backpropagation(BP)algoritim is very slow inconvergence speed in the high-dimensional weight space of neural network, the exact linesearches required by conjugate gradient algorithm is very difficult in implementation. This paperpresents a novel BP algorithm, which greatly improves leaming speed by using a modifiedconjugate gradient algorithm incorporating a simple inexact line sedch algorithm. Given manytesting results show that the efficiency of the algorithm relative to conventional BP is of morethan two order magnitude.