An Adaptive Empirical Bayesian Method for Sparse Deep Learning
This work addresses the need for efficient and robust deep learning models, particularly in compression and adversarial defense, though it appears incremental as it builds on existing Bayesian and sparse methods.
The authors tackled the problem of sparse deep learning by proposing an adaptive empirical Bayesian method with self-adaptive spike-and-slab priors, achieving state-of-the-art performance on MNIST, Fashion MNIST, and CIFAR10 datasets, and improving resistance to adversarial attacks.
We propose a novel adaptive empirical Bayesian method for sparse deep learning, where the sparsity is ensured via a class of self-adaptive spike-and-slab priors. The proposed method works by alternatively sampling from an adaptive hierarchical posterior distribution using stochastic gradient Markov Chain Monte Carlo (MCMC) and smoothly optimizing the hyperparameters using stochastic approximation (SA). We further prove the convergence of the proposed method to the asymptotically correct distribution under mild conditions. Empirical applications of the proposed method lead to the state-of-the-art performance on MNIST and Fashion MNIST with shallow convolutional neural networks and the state-of-the-art compression performance on CIFAR10 with Residual Networks. The proposed method also improves resistance to adversarial attacks.