Comparing Rewinding and Fine-tuning in Neural Network Pruning
This work addresses the retraining step in neural network pruning, offering improved methods for practitioners, though it is incremental as it builds on existing pruning frameworks.
The paper compares fine-tuning with two rewinding techniques for retraining pruned neural networks, finding that both rewinding methods outperform fine-tuning and achieve accuracy and compression ratios comparable to network-specific state-of-the-art methods.
Many neural network pruning algorithms proceed in three steps: train the network to completion, remove unwanted structure to compress the network, and retrain the remaining structure to recover lost accuracy. The standard retraining technique, fine-tuning, trains the unpruned weights from their final trained values using a small fixed learning rate. In this paper, we compare fine-tuning to alternative retraining techniques. Weight rewinding (as proposed by Frankle et al., (2019)), rewinds unpruned weights to their values from earlier in training and retrains them from there using the original training schedule. Learning rate rewinding (which we propose) trains the unpruned weights from their final values using the same learning rate schedule as weight rewinding. Both rewinding techniques outperform fine-tuning, forming the basis of a network-agnostic pruning algorithm that matches the accuracy and compression ratios of several more network-specific state-of-the-art techniques.