In this paper are presents and investigates the various modifications of gradient descent method for Back-Propagation algorithm. Opportunity on reduction of neural network training time and increasing of training algorithm stability are considered The particular expressions for choosing of optimum ...