Sparse Training via Boosting Pruning Plasticity with Neuroregeneration

Abstract

Works on lottery ticket hypothesis (LTH) and single-shot network pruning (SNIP) have raised a lot of attention currently on post-training pruning (iterative magnitude pruning), and before-training pruning (pruning at initialization). The former method suffers from an extremely large computation cost and the latter usually struggles with insufficient performance. In comparison, during-training pruning, a class of pruning methods that simultaneously enjoys the training/inference efficiency and the comparable performance, temporarily, has been less explored. To better understand during-training pruning, we quantitatively study the effect of pruning throughout training from the perspective of pruning plasticity (the ability of the pruned networks to recover the original performance). Pruning plasticity can help explain several other empirical observations about neural network pruning in literature. We further find that pruning plasticity can be substantially improved by injecting a brain-inspired mechanism called neuroregeneration, i.e., to regenerate the same number of connections as pruned. We design a novel gradual magnitude pruning (GMP) method, named gradual pruning with zero-cost neuroregeneration (\textbf{GraNet}), that advances state of the art. Perhaps most impressively, its sparse-to-sparse version for the first time boosts the sparse-to-sparse training performance over various dense-to-sparse methods with ResNet-50 on ImageNet without extending the training time. We release all codes in https://github.com/Shiweiliuiiiiiii/GraNet.

Published on the thirty-fifth Conference on Neural Information Processing Systems (NeurIPS 2021). Code can be found https://github.com/Shiweiliuiiiiiii/GraNet

References in corpus (21)

• To prune, or not to prune: exploring the efficacy of pruning for model compression
• Speeding up Convolutional Neural Networks with Low Rank Expansions
• The State of Sparsity in Deep Neural Networks
• Comparing Rewinding and Fine-tuning in Neural Network Pruning
• Picking Winning Tickets Before Training by Preserving Gradient Flow
• Carbon Emissions and Large Neural Network Training
• Parameter Efficient Training of Deep Convolutional Neural Networks by Dynamic Sparse Reparameterization
• Soft Threshold Weight Reparameterization for Learnable Sparsity
• Dynamic Model Pruning with Feedback
• Learning N:M Fine-grained Structured Sparse Neural Networks From Scratch
• Fixing the train-test resolution discrepancy: FixEfficientNet
• Winning the Lottery with Continuous Sparsification
• Dynamic Sparse Training: Find Efficient Sparse Network From Scratch With Trainable Masked Layers
• Top-KAST: Top-K Always Sparse Training
• Do We Actually Need Dense Over-Parameterization? In-Time Over-Parameterization in Sparse Training
• Selfish Sparse RNN Training
• Accelerated Sparse Neural Training: A Provable and Efficient Method to Find N:M Transposable Masks
• Trained Rank Pruning for Efficient Deep Neural Networks
• Pruning via Iterative Ranking of Sensitivity Statistics
• A Unified Paths Perspective for Pruning at Initialization
• CPOT: Channel Pruning via Optimal Transport

Cited by in corpus (1)

• Efficient Neural Network Training via Forward and Backward Propagation Sparsification