How Well Do Sparse Imagenet Models Transfer?

Abstract

Transfer learning is a classic paradigm by which models pretrained on large "upstream" datasets are adapted to yield good results on "downstream" specialized datasets. Generally, more accurate models on the "upstream" dataset tend to provide better transfer accuracy "downstream". In this work, we perform an in-depth investigation of this phenomenon in the context of convolutional neural networks (CNNs) trained on the ImageNet dataset, which have been pruned - that is, compressed by sparsifying their connections. We consider transfer using unstructured pruned models obtained by applying several state-of-the-art pruning methods, including magnitude-based, second-order, re-growth, lottery-ticket, and regularization approaches, in the context of twelve standard transfer tasks. In a nutshell, our study shows that sparse models can match or even outperform the transfer performance of dense models, even at high sparsities, and, while doing so, can lead to significant inference and even training speedups. At the same time, we observe and analyze significant differences in the behaviour of different pruning methods.

arXiv commentAccepted to CVPR'22. This version: 25 pages, 9 figures (including appendix). **Includes extended upstream training results, which are not present in the CVPR version.**

References in corpus (21)

• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
• YOLOv3: An Incremental Improvement
• Pruning Convolutional Neural Networks for Resource Efficient Inference
• Fine-Grained Visual Classification of Aircraft
• To prune, or not to prune: exploring the efficacy of pruning for model compression
• The State of Sparsity in Deep Neural Networks
• Do ImageNet Classifiers Generalize to ImageNet?
• Variational Dropout Sparsifies Deep Neural Networks
• Stabilizing the Lottery Ticket Hypothesis
• When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations
• Soft Threshold Weight Reparameterization for Learnable Sparsity
• Dynamic Model Pruning with Feedback
• Are we done with ImageNet?
• Training Skinny Deep Neural Networks with Iterative Hard Thresholding Methods
• Movement Pruning: Adaptive Sparsity by Fine-Tuning
• EigenDamage: Structured Pruning in the Kronecker-Factored Eigenbasis
• Top-KAST: Top-K Always Sparse Training
• Sparse Transfer Learning via Winning Lottery Tickets
• Accelerating Sparse Deep Neural Networks
• DeepHoyer: Learning Sparser Neural Network with Differentiable Scale-Invariant Sparsity Measures
• AC/DC: Alternating Compressed/DeCompressed Training of Deep Neural Networks