Exponential convergence rates for Batch Normalization: The power of length-direction decoupling in non-convex optimization

Jonas Kohler, Hadi Daneshmand, Aurelien Lucchi, Thomas Hofmann, Ming Zhou, Klaus Neymeyr
Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, PMLR 89:806-815, 2019.

Abstract

Normalization techniques such as Batch Normalization have been applied very successfully for training deep neural networks. Yet, despite its apparent empirical benefits, the reasons behind the success of Batch Normalization are mostly hypothetical. We here aim to provide a more thorough theoretical understanding from a classical optimization perspective. Our main contribution towards this goal is the identification of various problem instances in the realm of machine learning where Batch Normalization can provably accelerate optimization. We argue that this acceleration is due to the fact that Batch Normalization splits the optimization task into optimizing length and direction of the parameters separately. This allows gradient-based methods to leverage a favourable global structure in the loss landscape that we prove to exist in Learning Halfspace problems and neural network training with Gaussian inputs. We thereby turn Batch Normalization from an effective practical heuristic into a provably converging algorithm for these settings. Furthermore, we substantiate our analysis with empirical evidence that suggests the validity of our theoretical results in a broader context.

Cite this Paper


BibTeX
@InProceedings{pmlr-v89-kohler19a, title = {Exponential convergence rates for Batch Normalization: The power of length-direction decoupling in non-convex optimization}, author = {Kohler, Jonas and Daneshmand, Hadi and Lucchi, Aurelien and Hofmann, Thomas and Zhou, Ming and Neymeyr, Klaus}, booktitle = {Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics}, pages = {806--815}, year = {2019}, editor = {Chaudhuri, Kamalika and Sugiyama, Masashi}, volume = {89}, series = {Proceedings of Machine Learning Research}, month = {16--18 Apr}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v89/kohler19a/kohler19a.pdf}, url = {http://proceedings.mlr.press/v89/kohler19a.html}, abstract = {Normalization techniques such as Batch Normalization have been applied very successfully for training deep neural networks. Yet, despite its apparent empirical benefits, the reasons behind the success of Batch Normalization are mostly hypothetical. We here aim to provide a more thorough theoretical understanding from a classical optimization perspective. Our main contribution towards this goal is the identification of various problem instances in the realm of machine learning where Batch Normalization can provably accelerate optimization. We argue that this acceleration is due to the fact that Batch Normalization splits the optimization task into optimizing length and direction of the parameters separately. This allows gradient-based methods to leverage a favourable global structure in the loss landscape that we prove to exist in Learning Halfspace problems and neural network training with Gaussian inputs. We thereby turn Batch Normalization from an effective practical heuristic into a provably converging algorithm for these settings. Furthermore, we substantiate our analysis with empirical evidence that suggests the validity of our theoretical results in a broader context.} }
Endnote
%0 Conference Paper %T Exponential convergence rates for Batch Normalization: The power of length-direction decoupling in non-convex optimization %A Jonas Kohler %A Hadi Daneshmand %A Aurelien Lucchi %A Thomas Hofmann %A Ming Zhou %A Klaus Neymeyr %B Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Masashi Sugiyama %F pmlr-v89-kohler19a %I PMLR %P 806--815 %U http://proceedings.mlr.press/v89/kohler19a.html %V 89 %X Normalization techniques such as Batch Normalization have been applied very successfully for training deep neural networks. Yet, despite its apparent empirical benefits, the reasons behind the success of Batch Normalization are mostly hypothetical. We here aim to provide a more thorough theoretical understanding from a classical optimization perspective. Our main contribution towards this goal is the identification of various problem instances in the realm of machine learning where Batch Normalization can provably accelerate optimization. We argue that this acceleration is due to the fact that Batch Normalization splits the optimization task into optimizing length and direction of the parameters separately. This allows gradient-based methods to leverage a favourable global structure in the loss landscape that we prove to exist in Learning Halfspace problems and neural network training with Gaussian inputs. We thereby turn Batch Normalization from an effective practical heuristic into a provably converging algorithm for these settings. Furthermore, we substantiate our analysis with empirical evidence that suggests the validity of our theoretical results in a broader context.
APA
Kohler, J., Daneshmand, H., Lucchi, A., Hofmann, T., Zhou, M. & Neymeyr, K.. (2019). Exponential convergence rates for Batch Normalization: The power of length-direction decoupling in non-convex optimization. Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 89:806-815 Available from http://proceedings.mlr.press/v89/kohler19a.html.

Related Material