A Modified RMFI Conjugate Gradient-based Recurrent Neural Network

This paper studies the conjugate gradient algorithm combining generalized Armijo search technique based on recurrent neural network model,the main content of the thesis is presented as follows:In Chapter 1,the author briefly introduces the development status of conjugate gradient method and the research background of neural network,and summarizes the current research status of algorithms based on recurrent neural network model.In Chapter 2,a modified conjugate gradient algorithm is presented by combining with the generalized Armijo search strategy,based on Elman recurrent network model.The new algorithm modifies the RMFI conjugate gradient algorithm and updates the conjugate parameters,so that the algorithm can generate a sufficiently decreasing search direction for each iteration.In addition,the global convergence of the algorithm based on Elman model is proved.The experimental results show the effectiveness of the new algorithm.In Chapter 3,the RMFI conjugate gradient algorithm is improved,using a modified RMFI conjugate parameter.In particular,the calculation formula of the conjugate direction is improved,and the new algorithm produces a sufficient descent direction.The global convergence of the new algorithm under the generalized Armijo search technique is rigorously proved.Numerical experiments show that the new algorithm is superior to other algorithms.