PruneForget: Joint Unlearning and Pruning of Vision Models

2026-09-26

AI Summary

This study addresses the inefficiency and objective conflicts arising from treating machine unlearning and network pruning as independent processes in vision models. We propose a joint unlearning and pruning framework that, for the first time, reveals the intrinsic consistency between these two tasks. Specifically, our method leverages the forget set to guide parameter removal, replacing conventional serial pipelines with a mutually beneficial mechanism. Furthermore, we design a gradient-based joint optimization algorithm that maps information from unlearned samples into pruning weights. Experimental results demonstrate that this framework effectively eliminates the influence of specific data across both classifiers and generative models while substantially reducing inference costs and memory footprints. Notably, its performance closely approximates the ideal baseline of post-retraining pruning, thereby achieving simultaneous improvements in safety compliance and resource efficiency.

Abstract

Machine unlearning and model pruning are increasingly coupled in the real world. Models must support unlearning requests, e.g., for safety concerns, while also meeting requirements in latency and memory budget. Until recently, existing works have studied each aspect as an independent problem, e.g., running unlearning and pruning sequentially. In this work, we show that unlearning and pruning are naturally aligned and should be solved jointly to be made aware of each other. Intuitively, parameters that encode information of the unlearned samples are natural pruning targets, as unlearning and pruning both call for the"deletion"of such parameters. We propose PruneForget, a method that uses the unlearn set as a guide for pruning, so that unlearning and pruning mutually benefit each other. Extensive experiments on image classifiers and generative models show that PruneForget removes the influence of the unlearned samples while producing a more compact model with reduced inference cost. It achieves a negligible performance gap relative to an oracle that retrains from scratch for unlearning and then prunes.

Problem

• machine unlearning
• model pruning
• vision models
• joint optimization

Innovation

• Machine Unlearning
• Model Pruning
• Joint Optimization
• Vision Models
• PruneForget