DessiLBI: Exploring Structural Sparsity of Deep Networks via Differential Inclusion Paths

Yanwei Fu, Chen Liu, Donghao Li, Xinwei Sun, Jinshan Zeng, Yuan Yao
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:3315-3326, 2020.

Abstract

Over-parameterization is ubiquitous nowadays in training neural networks to benefit both optimization in seeking global optima and generalization in reducing prediction error. However, compressive networks are desired in many real world applications and direct training of small networks may be trapped in local optima. In this paper, instead of pruning or distilling over-parameterized models to compressive ones, we propose a new approach based on differential inclusions of inverse scale spaces. Specifically, it generates a family of models from simple to complex ones that couples a pair of parameters to simultaneously train over-parameterized deep models and structural sparsity on weights of fully connected and convolutional layers. Such a differential inclusion scheme has a simple discretization, proposed as Deep structurally splitting Linearized Bregman Iteration (DessiLBI), whose global convergence analysis in deep learning is established that from any initializations, algorithmic iterations converge to a critical point of empirical risks. Experimental evidence shows that DessiLBI achieve comparable and even better performance than the competitive optimizers in exploring the structural sparsity of several widely used backbones on the benchmark datasets. Remarkably, with early stopping, DessiLBI unveils “winning tickets” in early epochs: the effective sparse structure with comparable test accuracy to fully trained over-parameterized models.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-fu20d, title = {{D}essi{LBI}: Exploring Structural Sparsity of Deep Networks via Differential Inclusion Paths}, author = {Fu, Yanwei and Liu, Chen and Li, Donghao and Sun, Xinwei and Zeng, Jinshan and Yao, Yuan}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {3315--3326}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/fu20d/fu20d.pdf}, url = {https://proceedings.mlr.press/v119/fu20d.html}, abstract = {Over-parameterization is ubiquitous nowadays in training neural networks to benefit both optimization in seeking global optima and generalization in reducing prediction error. However, compressive networks are desired in many real world applications and direct training of small networks may be trapped in local optima. In this paper, instead of pruning or distilling over-parameterized models to compressive ones, we propose a new approach based on differential inclusions of inverse scale spaces. Specifically, it generates a family of models from simple to complex ones that couples a pair of parameters to simultaneously train over-parameterized deep models and structural sparsity on weights of fully connected and convolutional layers. Such a differential inclusion scheme has a simple discretization, proposed as Deep structurally splitting Linearized Bregman Iteration (DessiLBI), whose global convergence analysis in deep learning is established that from any initializations, algorithmic iterations converge to a critical point of empirical risks. Experimental evidence shows that DessiLBI achieve comparable and even better performance than the competitive optimizers in exploring the structural sparsity of several widely used backbones on the benchmark datasets. Remarkably, with early stopping, DessiLBI unveils “winning tickets” in early epochs: the effective sparse structure with comparable test accuracy to fully trained over-parameterized models.} }
Endnote
%0 Conference Paper %T DessiLBI: Exploring Structural Sparsity of Deep Networks via Differential Inclusion Paths %A Yanwei Fu %A Chen Liu %A Donghao Li %A Xinwei Sun %A Jinshan Zeng %A Yuan Yao %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-fu20d %I PMLR %P 3315--3326 %U https://proceedings.mlr.press/v119/fu20d.html %V 119 %X Over-parameterization is ubiquitous nowadays in training neural networks to benefit both optimization in seeking global optima and generalization in reducing prediction error. However, compressive networks are desired in many real world applications and direct training of small networks may be trapped in local optima. In this paper, instead of pruning or distilling over-parameterized models to compressive ones, we propose a new approach based on differential inclusions of inverse scale spaces. Specifically, it generates a family of models from simple to complex ones that couples a pair of parameters to simultaneously train over-parameterized deep models and structural sparsity on weights of fully connected and convolutional layers. Such a differential inclusion scheme has a simple discretization, proposed as Deep structurally splitting Linearized Bregman Iteration (DessiLBI), whose global convergence analysis in deep learning is established that from any initializations, algorithmic iterations converge to a critical point of empirical risks. Experimental evidence shows that DessiLBI achieve comparable and even better performance than the competitive optimizers in exploring the structural sparsity of several widely used backbones on the benchmark datasets. Remarkably, with early stopping, DessiLBI unveils “winning tickets” in early epochs: the effective sparse structure with comparable test accuracy to fully trained over-parameterized models.
APA
Fu, Y., Liu, C., Li, D., Sun, X., Zeng, J. & Yao, Y.. (2020). DessiLBI: Exploring Structural Sparsity of Deep Networks via Differential Inclusion Paths. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:3315-3326 Available from https://proceedings.mlr.press/v119/fu20d.html.

Related Material