Growing Efficient Deep Networks by Structured Continuous Sparsification
This addresses the computational expense of training deep networks for researchers and practitioners, offering a novel approach that is not incremental but provides broad efficiency gains.
The paper tackles the problem of training deep networks efficiently by developing a method that grows and prunes architectures during training, achieving 49.7% inference FLOPs savings and 47.4% training FLOPs savings on ImageNet while maintaining 75.2% top-1 accuracy without fine-tuning.
We develop an approach to growing deep network architectures over the course of training, driven by a principled combination of accuracy and sparsity objectives. Unlike existing pruning or architecture search techniques that operate on full-sized models or supernet architectures, our method can start from a small, simple seed architecture and dynamically grow and prune both layers and filters. By combining a continuous relaxation of discrete network structure optimization with a scheme for sampling sparse subnetworks, we produce compact, pruned networks, while also drastically reducing the computational expense of training. For example, we achieve $49.7\%$ inference FLOPs and $47.4\%$ training FLOPs savings compared to a baseline ResNet-50 on ImageNet, while maintaining $75.2\%$ top-1 accuracy -- all without any dedicated fine-tuning stage. Experiments across CIFAR, ImageNet, PASCAL VOC, and Penn Treebank, with convolutional networks for image classification and semantic segmentation, and recurrent networks for language modeling, demonstrate that we both train faster and produce more efficient networks than competing architecture pruning or search methods.