Training feed forward neural network with modified Fletcher-Reeves method

Abstract

In this research, a modified Fletcher-Reeves (FR) conjugate gradient algorithm for training large scale feed forward neural network (FFNN) is presented. Under mild conditions, we establish that the proposed method satisfies the sufficient descent condition, and it is globally convergent under wolfe line search condition. The evidence which is provided by experimental results showed that our proposed method is preferable and superior to the classic methods.