Deadwooding: Robust Global Pruning for Deep Neural Networks

TL;DR

Deadwooding introduces a global pruning method for deep neural networks that enhances sparsity, robustness, and accuracy simultaneously, addressing limitations of existing pruning techniques in resource-limited deployment scenarios.

Contribution

The paper presents Deadwooding, a novel Lagrangian Dual-based global pruning approach that improves model sparsity, robustness, and accuracy over prior methods.

Findings

Outperforms state-of-the-art in robustness and accuracy

Effectively balances sparsity and model performance

Enhances deployment feasibility in resource-constrained environments

Abstract

The ability of Deep Neural Networks to approximate highly complex functions is key to their success. This benefit, however, comes at the expense of a large model size, which challenges its deployment in resource-constrained environments. Pruning is an effective technique used to limit this issue, but often comes at the cost of reduced accuracy and adversarial robustness. This paper addresses these shortcomings and introduces Deadwooding, a novel global pruning technique that exploits a Lagrangian Dual method to encourage model sparsity while retaining accuracy and ensuring robustness. The resulting model is shown to significantly outperform the state-of-the-art studies in measures of robustness and accuracy.