Optimal Structure of Pipelined Recurrent Neural Network in Modeling of MPEG Video

This paper investigates optimal structure of Piplined Recurrent Neural Network (PRNN) for adaptive traffic prediction of MPEG video signal via dynamic ATM networks. The traffic signal of each picture type (I, P, and B) of MPEG video is characterized by a nonlinear autoregressive moving average (NARMA) process. Since those modules of PRNN can be performed simultaneously in a pipelined parallelism fashion, this would lead to a significant improvement in the total computational efficiency of PRNN. In order to further improve the convergence performance of the adaptive algorithm for PRNN, a learning-rate annealing schedule is proposed to accelerate the adaptive learning process. The measure that is used for optimum structure of PRNN i.e. number of neurons in each module, number of modules is the one-step forward prediction gain. Results are shown to be promising and practically feasible in obtaining the best structure for a adaptive predictor of Noise.