The Flip Side of the Reweighted Coin: Duality of Adaptive Dropout and Regularization

Abstract

Among the most successful methods for sparsifying deep (neural) networks are those that adaptively mask the network weights throughout training. By examining this masking, or dropout, in the linear case, we uncover a duality between such adaptive methods and regularization through the so-called "$\eta$-trick" that casts both as iteratively reweighted optimizations. We show that any dropout strategy that adapts to the weights in a monotonic way corresponds to an effective subquadratic regularization penalty, and therefore leads to sparse solutions. We obtain the effective penalties for several popular sparsification strategies, which are remarkably similar to classical penalties commonly used in sparse optimization. Considering variational dropout as a case study, we demonstrate similar empirical behavior between the adaptive dropout method and classical methods on the task of deep network sparsification, validating our theory.

19 pages, 2 figures. Appeared in NeurIPS 2021. Small typographical correction

References in corpus (7)

• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Nearly unbiased variable selection under minimax concave penalty
• To prune, or not to prune: exploring the efficacy of pruning for model compression
• The State of Sparsity in Deep Neural Networks
• Structured Sparse Principal Component Analysis
• On Convergence and Generalization of Dropout Training
• Dropout: Explicit Forms and Capacity Control