Donghao Li

Yanwei Fu, Chen Liu, Donghao Li et al.

The great success of deep neural networks is built upon their over-parameterization, which smooths the optimization landscape without degrading the generalization ability. Despite the benefits of over-parameterization, a huge amount of parameters makes deep networks cumbersome in daily life applications. Though techniques such as pruning and distillation are developed, they are expensive in fully training a dense network as backward selection methods, and there is still a void on systematically exploring forward selection methods for learning structural sparsity in deep networks. To fill in this gap, this paper proposes a new approach based on differential inclusions of inverse scale spaces, which generate a family of models from simple to complex ones along the dynamics via coupling a pair of parameters, such that over-parameterized deep models and their structural sparsity can be explored simultaneously. This kind of differential inclusion scheme has a simple discretization, dubbed Deep structure splitting Linearized Bregman Iteration (DessiLBI), whose global convergence in learning deep networks could be established under the Kurdyka-Lojasiewicz framework. Experimental evidence shows that our method achieves comparable and even better performance than the competitive optimizers in exploring the sparse structure of several widely used backbones on the benchmark datasets. Remarkably, with early stopping, our method unveils `winning tickets' in early epochs: the effective sparse network structures with comparable test accuracy to fully trained over-parameterized models, that are further transferable to similar alternative tasks. Furthermore, our method is able to grow networks efficiently with adaptive filter configurations, demonstrating a good performance with much less computational cost. Codes and models can be downloaded at {https://github.com/DessiLBI2020/DessiLBI}.

Yanwei Fu, Chen Liu, Donghao Li et al.

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.

Yanwei Fu, Donghao Li, Xinwei Sun et al.

This paper proposes a novel Stochastic Split Linearized Bregman Iteration ($S^{2}$-LBI) algorithm to efficiently train the deep network. The $S^{2}$-LBI introduces an iterative regularization path with structural sparsity. Our $S^{2}$-LBI combines the computational efficiency of the LBI, and model selection consistency in learning the structural sparsity. The computed solution path intrinsically enables us to enlarge or simplify a network, which theoretically, is benefited from the dynamics property of our $S^{2}$-LBI algorithm. The experimental results validate our $S^{2}$-LBI on MNIST and CIFAR-10 dataset. For example, in MNIST, we can either boost a network with only 1.5K parameters (1 convolutional layer of 5 filters, and 1 FC layer), achieves 98.40\% recognition accuracy; or we simplify $82.5\%$ of parameters in LeNet-5 network, and still achieves the 98.47\% recognition accuracy. In addition, we also have the learning results on ImageNet, which will be added in the next version of our report.