Masked Training of Neural Networks with Partial Gradients

Abstract

State-of-the-art training algorithms for deep learning models are based on stochastic gradient descent (SGD). Recently, many variations have been explored: perturbing parameters for better accuracy (such as in Extragradient), limiting SGD updates to a subset of parameters for increased efficiency (such as meProp) or a combination of both (such as Dropout). However, the convergence of these methods is often not studied in theory. We propose a unified theoretical framework to study such SGD variants -- encompassing the aforementioned algorithms and additionally a broad variety of methods used for communication efficient training or model compression. Our insights can be used as a guide to improve the efficiency of such methods and facilitate generalization to new applications. As an example, we tackle the task of jointly training networks, a version of which (limited to sub-networks) is used to create Slimmable Networks. By training a low-rank Transformer jointly with a standard one we obtain superior performance than when it is trained separately.

Proceedings of the 25th International Conference on Artificial Intelligence and Statistics (AISTATS) 2022

References in corpus (9)

• Distilling the Knowledge in a Neural Network
• To prune, or not to prune: exploring the efficacy of pruning for model compression
• AutoSlim: Towards One-Shot Architecture Search for Channel Numbers
• Dynamic Model Pruning with Feedback
• Training Skinny Deep Neural Networks with Iterative Hard Thresholding Methods
• Characterising Bias in Compressed Models
• On the Convergence of SGD with Biased Gradients
• AC/DC: Alternating Compressed/DeCompressed Training of Deep Neural Networks
• Pufferfish: Communication-efficient Models At No Extra Cost