AISTATS 2022poster32 citations

Masked Training of Neural Networks with Partial Gradients

Amirkeivan Mohtashami, Martin Jaggi, Sebastian Stich

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.

BibTeX
@InProceedings{pmlr-v151-mohtashami22a,
  title = 	 { Masked Training of Neural Networks with Partial Gradients },
  author =       {Mohtashami, Amirkeivan and Jaggi, Martin and Stich, Sebastian},
  booktitle = 	 {Proceedings of The 25th International Conference on Artificial Intelligence and Statistics},
  pages = 	 {5876--5890},
  year = 	 {2022},
  editor = 	 {Camps-Valls, Gustau and Ruiz, Francisco J. R. and Valera, Isabel},
  volume = 	 {151},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {28--30 Mar},
  publisher =    {PMLR},
  pdf = 	 {https://proceedings.mlr.press/v151/mohtashami22a/mohtashami22a.pdf},
  url = 	 {https://proceedings.mlr.press/v151/mohtashami22a.html},
  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. }
}
Masked Training of Neural Networks with Partial Gradients · AISTATS 2022