Attention Smoothing Is All You Need For Unlearning

TL;DR

The paper introduces Attention Smoothing Unlearning (ASU), a novel method that effectively erases memorized knowledge in large language models by smoothing attention distributions, improving unlearning stability and utility retention.

Contribution

ASU is a new unlearning framework that uses self-distillation and attention smoothing to better erase memorized data while preserving model coherence.

Findings

ASU outperforms baselines in various unlearning scenarios.

ASU maintains model utility with minimal loss.

ASU effectively erases factual memorization.

Abstract

Large Language Models are prone to memorizing sensitive, copyrighted, or hazardous content, posing significant privacy and legal concerns. Retraining from scratch is computationally infeasible, whereas current unlearning methods exhibit unstable trade-offs between forgetting and utility, frequently producing incoherent outputs on forget prompts and failing to generalize due to the persistence of lexical-level and semantic-level associations in attention. We propose Attention Smoothing Unlearning (ASU), a principled framework that casts unlearning as self-distillation from a forget-teacher derived from the model's own attention. By increasing the softmax temperature, ASU flattens attention distributions and directly suppresses the lexical-level and semantic-level associations responsible for reconstructing memorized knowledge. This results in a bounded optimization objective that erases factual information yet maintains coherence in responses to forget prompts. Empirical evaluation on TOFU, MUSE, and WMDP, along with real-world and continual unlearning scenarios across question answering and text completion, demonstrates that ASU outperforms the baselines for most unlearning scenarios, delivering robust unlearning with minimal loss of model utility.