Improving post-training structured pruning via two-stage reconstruction

doi:10.1016/j.eswa.2025.128930

Improving post-training structured pruning via two-stage reconstruction

2025-07-05

0

PRE

AI

DOI：10.1016/j.eswa.2025.128930

Original

Original request for help

Save

Abstract

En 中文

• Structured pruning removes entire channels, and post-training calibration methods are difficult to restore accuracy. • The pre-reconstruction step aggregates information and reduces pruning damage. • Global output reconstruction modeling accumulated errors to better restore model accuracy. • Uses about 1 % and recovers the accuracy loss caused by pruning within minutes.

Improving post-training structured pruning via two-stage reconstruction

Abstract

Journal

Organization