Comparative study on various pruning algorithms for RNN. II. Experimental results and comparative analysis

For Pt. I see ibid., p.2225-30 (2002). Owing to the computational complexity requirement, pruning a fully connected recurrent neural network (RNN) would be ineffective for large size RNN. In Part I, several non-heuristic pruning algorithms for fully connected RNN have been proposed, some of them are extended from EKF based approaches and some of them are based on weight magnitude, together with some techniques on the pruning procedures. Their effectiveness on the computational complexities have been analyzed. In this paper, their effectiveness on network sizes and the generalization abilities will be evaluated.