One-Cycle Structured Pruning via Stability-Driven Subnetwork Search

TL;DR

This paper introduces a one-cycle structured pruning method that efficiently identifies optimal subnetworks during early training, reducing computational costs while maintaining high accuracy across multiple datasets and architectures.

Contribution

It presents a novel one-cycle pruning framework combining pre-training, pruning, and fine-tuning, guided by a stability-driven subnetwork search and regularization to improve efficiency.

Findings

Achieves state-of-the-art accuracy with reduced training cost.

Effectively identifies stable pruning epochs using a new indicator.

Demonstrates superior efficiency on CIFAR and ImageNet datasets.

Abstract

Existing structured pruning methods typically rely on multi-stage training procedures that incur high computational costs. Pruning at initialization aims to reduce this burden but often suffers from degraded performance. To address these limitations, we propose an efficient one-cycle structured pruning framework that integrates pre-training, pruning, and fine-tuning into a single training cycle without sacrificing accuracy. The key idea is to identify an optimal sub-network during the early stages of training, guided by norm-based group saliency criteria and structured sparsity regularization. We introduce a novel pruning indicator that detects a stable pruning epoch by measuring the similarity between pruning sub-networks across consecutive training epochs. In addition, group sparsity regularization accelerates convergence, further reducing overall training time. Extensive experiments on CIFAR-10, CIFAR-100, and ImageNet using VGG, ResNet, and MobileNet architectures demonstrate that the proposed method achieves state-of-the-art accuracy while being among the most efficient structured pruning frameworks in terms of training cost. Code is available at https://github.com/ghimiredhikura/OCSPruner.