Exploring Criteria of Loss Reweighting to Enhance LLM Unlearning

TL;DR

This paper investigates loss reweighting strategies to improve large language model unlearning, identifying two goals—Saturation and Importance—and proposing a combined method called SatImp that enhances unlearning effectiveness.

Contribution

The paper clarifies the roles of Saturation and Importance in loss reweighting, and introduces SatImp, a novel combined strategy validated through extensive empirical analysis.

Findings

Saturation-based reweighting outperforms importance-based methods.

Combining Saturation and Importance yields additional improvements.

The efficacy depends on the smoothness and granularity of weight distributions.

Abstract

Loss reweighting has shown significant benefits for machine unlearning with large language models (LLMs). However, their exact functionalities are left unclear and the optimal strategy remains an open question, thus impeding the understanding and improvement of existing methodologies. In this paper, we identify two distinct goals of loss reweighting, namely, Saturation and Importance -- the former indicates that those insufficiently optimized data should be emphasized, while the latter stresses some critical data that are most influential for loss minimization. To study their usefulness, we design specific reweighting strategies for each goal and evaluate their respective effects on unlearning. We conduct extensive empirical analyses on well-established benchmarks, and summarize some important observations as follows: (i) Saturation enhances efficacy more than importance-based reweighting, and their combination can yield additional improvements. (ii) Saturation typically allocates lower weights to data with lower likelihoods, whereas importance-based reweighting does the opposite. (iii) The efficacy of unlearning is also largely influenced by the smoothness and granularity of the weight distributions. Based on these findings, we propose SatImp, a simple reweighting method that combines the advantages of both saturation and importance. Empirical results on extensive datasets validate the efficacy of our method, potentially bridging existing research gaps and indicating directions for future research. Our code is available at https://github.com/tmlr-group/SatImp.