Progressive Skeletonization: Trimming more fat from a network at initialization

TL;DR

This paper introduces a new method called FORCE for network skeletonization at initialization, enabling extremely high pruning levels (up to 99.5%) while maintaining trainability and performance, surpassing existing approaches especially at high sparsity.

Contribution

The paper proposes the FORCE objective and two approximation procedures that improve network pruning at initialization, allowing for higher sparsity levels without performance degradation.

Findings

FORCE achieves up to 99.5% pruning while preserving trainability.

Compared to existing methods, FORCE performs better at high sparsity levels.

Empirical results demonstrate the effectiveness of the proposed approach.

Abstract

Recent studies have shown that skeletonization (pruning parameters) of networks \textit{at initialization} provides all the practical benefits of sparsity both at inference and training time, while only marginally degrading their performance. However, we observe that beyond a certain level of sparsity (approx $95\%$), these approaches fail to preserve the network performance, and to our surprise, in many cases perform even worse than trivial random pruning. To this end, we propose an objective to find a skeletonized network with maximum {\em foresight connection sensitivity} (FORCE) whereby the trainability, in terms of connection sensitivity, of a pruned network is taken into consideration. We then propose two approximate procedures to maximize our objective (1) Iterative SNIP: allows parameters that were unimportant at earlier stages of skeletonization to become important at later stages; and (2) FORCE: iterative process that allows exploration by allowing already pruned parameters to resurrect at later stages of skeletonization. Empirical analyses on a large suite of experiments show that our approach, while providing at least as good a performance as other recent approaches on moderate pruning levels, provides remarkably improved performance on higher pruning levels (could remove up to $99.5\%$ parameters while keeping the networks trainable). Code can be found in https://github.com/naver/force.