Escape from the speed down areas on training of backpropagation algorithm by using a jumping method

The backpropagation algorithm (BP) is one of the most widely used training algorithms for supervised learning in multilayered neural networks (NNs). BP has widely been used in many fields. However, the state of the speed down area on training by BP can often be observed. It often takes a long time for training using BP. In this paper, the authors propose the jumping method that escapes from a speed down area in BP training by making the network coefficients remarkable large. As a result, training using the authors' method escapes from the speed down area quickly.