Compressing Neural Networks using the Variational Information Bottleneck

Bin Dai, Chen Zhu, Baining Guo, David Wipf
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:1135-1144, 2018.

Abstract

Neural networks can be compressed to reduce memory and computational requirements, or to increase accuracy by facilitating the use of a larger base architecture. In this paper we focus on pruning individual neurons, which can simultaneously trim model size, FLOPs, and run-time memory. To improve upon the performance of existing compression algorithms we utilize the information bottleneck principle instantiated via a tractable variational bound. Minimization of this information theoretic bound reduces the redundancy between adjacent layers by aggregating useful information into a subset of neurons that can be preserved. In contrast, the activations of disposable neurons are shut off via an attractive form of sparse regularization that emerges naturally from this framework, providing tangible advantages over traditional sparsity penalties without contributing additional tuning parameters to the energy landscape. We demonstrate state-of-the-art compression rates across an array of datasets and network architectures.

Cite this Paper


BibTeX
@InProceedings{pmlr-v80-dai18d, title = {Compressing Neural Networks using the Variational Information Bottleneck}, author = {Dai, Bin and Zhu, Chen and Guo, Baining and Wipf, David}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {1135--1144}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/dai18d/dai18d.pdf}, url = {https://proceedings.mlr.press/v80/dai18d.html}, abstract = {Neural networks can be compressed to reduce memory and computational requirements, or to increase accuracy by facilitating the use of a larger base architecture. In this paper we focus on pruning individual neurons, which can simultaneously trim model size, FLOPs, and run-time memory. To improve upon the performance of existing compression algorithms we utilize the information bottleneck principle instantiated via a tractable variational bound. Minimization of this information theoretic bound reduces the redundancy between adjacent layers by aggregating useful information into a subset of neurons that can be preserved. In contrast, the activations of disposable neurons are shut off via an attractive form of sparse regularization that emerges naturally from this framework, providing tangible advantages over traditional sparsity penalties without contributing additional tuning parameters to the energy landscape. We demonstrate state-of-the-art compression rates across an array of datasets and network architectures.} }
Endnote
%0 Conference Paper %T Compressing Neural Networks using the Variational Information Bottleneck %A Bin Dai %A Chen Zhu %A Baining Guo %A David Wipf %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-dai18d %I PMLR %P 1135--1144 %U https://proceedings.mlr.press/v80/dai18d.html %V 80 %X Neural networks can be compressed to reduce memory and computational requirements, or to increase accuracy by facilitating the use of a larger base architecture. In this paper we focus on pruning individual neurons, which can simultaneously trim model size, FLOPs, and run-time memory. To improve upon the performance of existing compression algorithms we utilize the information bottleneck principle instantiated via a tractable variational bound. Minimization of this information theoretic bound reduces the redundancy between adjacent layers by aggregating useful information into a subset of neurons that can be preserved. In contrast, the activations of disposable neurons are shut off via an attractive form of sparse regularization that emerges naturally from this framework, providing tangible advantages over traditional sparsity penalties without contributing additional tuning parameters to the energy landscape. We demonstrate state-of-the-art compression rates across an array of datasets and network architectures.
APA
Dai, B., Zhu, C., Guo, B. & Wipf, D.. (2018). Compressing Neural Networks using the Variational Information Bottleneck. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:1135-1144 Available from https://proceedings.mlr.press/v80/dai18d.html.

Related Material