PTGVLR: fast MLP learning using parallel tangent gradient with variable learning rates

In this paper, we propose a MLP learning algorithm based on the parallel tangent gradient with modified variable learning rates, PTGVLR. Parallel tangent gradient uses parallel tangent deflecting direction instead of the momentum. Moreover, we use two separate and variable learning rates one for the gradient descent and the other for accelerating direction through parallel tangent. We test PTGVLR optimization method for optimizing a two dimensional Rosenbrock function and for learning of some well-known MLP problems, such as the parity generators and the encoders. Our investigations show that the proposed MLP learning algorithm, PTGVLR, is faster than similar adaptive learning methods.