Sparsely connected asynchronous reservoir computing network

Abstract

In order to solve the reservoir computing networkconstruction problem, a sparsely connected asynchronous neuronreservoir construction method is proposed. The method connectsseveral sub-reservoirs sequentially and sets lag links amongsub-reservoirs in order to handle input signals asynchronously insub-reservoirs, and further constitutes serial memory. In order toachieve efficient information transmission, sparse connections areused among sub-reservoirs. Experimental results show that theproposed method can effectively improve the memory capacity ofreservoir and it is easy to deal with long-term dependence problems.In addition, the proposed structure makes the reservoir produce moreabundant dynamic behavior and has better robustness to the initialparameters.