A New Learning Method for S-GCM

Recently, there have been many studies on artificial neural network models with nonequilibrium dynamics. For example, Ishii et al.’s model, which is an improvement version of Kaneko’s globally coupled map (GCM) model, is called globally coupled map using the symmetric map (SGCM). A new learning method for S-GCM is proposed in this paper. In the proposed method, we use modified saprse matrix for learning method. Both the theory analyses and computer simulation results show that the performance of S-GCM can be improved greatly by using the MIMS learning method. Our learning method named as More Iterate More Store (MIMS) learning. The method is like sparse method, with difference in sparse method. This method to recur the stored patterns and in result it will be dependent on the sequence of storing the patterns, on the other hand, primary patterns have more effect in creating the weight matrix in comparison to the patterns will be stored finally, it means they are recurred more and consequently they are stored and stick better in the memory. It seems this method of learning is more similar to the man’s way of learning, as the patterns which we repetition during time we will keep them in our long-term memory better.