Layer-Wise Network Compression Using Gaussian Mixture Model

Due to the large number of parameters and heavy computation, real-time operation deep learning in low-performance embedded board is still difficult. Network Pruning one effective methods reduce without additional network structure modification. However, conventional method prunes redundant up same rate for all layers. It may cause a bottleneck problem, which leads performance degradation, because minimum optimal different according each layer. We propose layer adaptive pruning based on modeling weight distribution. can measure amount weights close zero accurately by applying Gaussian Mixture Model (GMM). Until target compression reached, selection are iteratively performed. The iteration considers timing reach degree pruning. apply proposed image classification semantic segmentation show effectiveness method. In experiments, shows higher during maintaining accuracy compared with previous methods.