Unlearning with Fisher Masking

TL;DR

This paper introduces Fisher Masking, a novel unlearning method that effectively removes specific training data from models by masking important parameters based on Fisher information, outperforming traditional fine-tuning approaches.

Contribution

The paper proposes a Fisher information-based masking strategy that enhances unlearning efficiency and stability without extensive fine-tuning.

Findings

Fisher Masking achieves near-complete unlearning without fine-tuning.

The method maintains high performance on remaining data.

It demonstrates superior stability compared to existing unlearning methods.

Abstract

Machine unlearning aims to revoke some training data after learning in response to requests from users, model developers, and administrators. Most previous methods are based on direct fine-tuning, which may neither remove data completely nor retain full performances on the remain data. In this work, we find that, by first masking some important parameters before fine-tuning, the performances of unlearning could be significantly improved. We propose a new masking strategy tailored to unlearning based on Fisher information. Experiments on various datasets and network structures show the effectiveness of the method: without any fine-tuning, the proposed Fisher masking could unlearn almost completely while maintaining most of the performance on the remain data. It also exhibits stronger stability compared to other unlearning baselines