Multilayer neural networks training methodic

We propose three new techniques for training of multilayer neural networks. Its basic concept is based on the gradient descent method. For every methodic are showed formulas for calculation of the adaptive training steps. Matrix algorithmization for all of this techniques are very helpful in its program realization.