Efficient layer-wise learning of feedforward neural networks using the backpropagation

In this paper, we propose the efficient method for learning a feedforward neural network (FNN) that combines the backpropagation (BP) strategy and the layer-wise learning methods (LWMs). More precisely, for updating weights of each layer we use the BP-based iterative procedure. This procedure provides the realistic conditions for fast convergence of the learning process to a global minima. As a result, the new method uses advantages and overcomes the disadvantages associated with both the BP method and LWMs.