Structured Pruning of Convolutional Neural Networks via L1 Regularization

Deep learning architecture has achieved amazing success in many areas with the recent advancements in convolutional neural networks (CNNs). However, real-time applications of CNNs are seriously hindered by the significant storage and computational costs. Structured pruning is a promising method to compress and accelerate CNNs and does not need special hardware or software for an auxiliary calculation. Here a simple strategy of structured pruning approach is proposed to crop unimportant filters or neurons automatically during the training stage. The proposed method introduces a mask for all filters or neurons to evaluate their importance. Thus the filters or neurons with zero mask are removed. To achieve this, the proposed method adopted L1 regularization to zero filters or neurons of CNNs. Experiments were conducted to assess the validity of this technique. The experiments showed that the proposed approach could crop 90.4%, 95.6% and 34.04% parameters on LeNet-5, VGG-16, and ResNet-32respectively, with a negligible loss of accuracy.