Hybrid algorithm training forward neural networks: Improving convergence speed and performance accuracy

This goal of this study to get better conjugate gradient algorithms (C.G. As) used to training feed-forward neural networks (F.F.N.N). This hybrid algorithm focused on the Fletcher (C.D) 1987 and Dai-Liao (D.L) 2001 algorithms, where adaptabilities to these algorithms are given to achieve more active performance and speediest convergence. The basic algorithms depend on C.G. As to solve unconstrained optimization problems, but they hardship of problems as slow convergence, also need for large steps to get the solution, and big storage capacity. This paper scout how to combining C.D and D.L algorithms with each other to evolve a new research direction by which the speed convergence and accuracy for solutions can improvement. Availability an improved alternative to normal algorithms depends on linear convex function; to experiment a hybrid algorithm it has been application on four types of real data. The results improved a hybrid algorithm has speediest converge, and high accuracy when compared with C.D and D.L algorithms.