Nonmonotone learning of recurrent neural networks in symbolic sequence processing applications

摘要: In this paper, we present a formulation of the learning problem that allows deterministic nonmonotone behaviour to be generated, i.e. values error function are allowed increase temporarily although is progressively improved. This achieved by introducing strategy on values. We four training algorithms which equipped with and investigate their performance in symbolic sequence processing problems. Experimental results show mechanism can improve traditional strategies make them more effective problems tested.