earticle

영어

This paper presents a recurrent neural network model which takes outputs of a hidden layer of multilayer neural network (MNN) as inputs. The proposed network model has a context layer for taking input streams out of the hidden layer. We have proved that the proposed recurrent neural network (RNN) model should be a generalized form of nonlinear autoregressive moving average (NARMA) model. We have tested performances of the proposed method for a few time series data. The prediction performances of the recurrent neural network model are shown to be competitive enough for a general predictor.