Operation-Aware Soft Channel Pruning using Differentiable Masks

Abstract

We propose a simple but effective data-driven channel pruning algorithm, which compresses deep neural networks in a differentiable way by exploiting the characteristics of operations. The proposed approach makes a joint consideration of batch normalization (BN) and rectified linear unit (ReLU) for channel pruning; it estimates how likely the two successive operations deactivate each feature map and prunes the channels with high probabilities. To this end, we learn differentiable masks for individual channels and make soft decisions throughout the optimization procedure, which facilitates to explore larger search space and train more stable networks. The proposed framework enables us to identify compressed models via a joint learning of model parameters and channel pruning without an extra procedure of fine-tuning. We perform extensive experiments and achieve outstanding performance in terms of the accuracy of output networks given the same amount of resources when compared with the state-of-the-art methods.

ICML 2020

Cited by in corpus (10)

• Only Train Once: A One-Shot Neural Network Training And Pruning Framework
• Accelerate CNNs from Three Dimensions: A Comprehensive Pruning Framework
• Efficient Neural Network Training via Forward and Backward Propagation Sparsification
• Convolutional Neural Network Pruning with Structural Redundancy Reduction
• Training Compact CNNs for Image Classification using Dynamic-coded Filter Fusion
• Effective Sparsification of Neural Networks with Global Sparsity Constraint
• FAT: Learning Low-Bitwidth Parametric Representation via Frequency-Aware Transformation
• Automatic Neural Network Pruning that Efficiently Preserves the Model Accuracy
• BN-NAS: Neural Architecture Search with Batch Normalization
• Carrying out CNN Channel Pruning in a White Box