Not Every Token Needs Forgetting: Selective Unlearning to Limit Change in Utility in Large Language Model Unlearning

TL;DR

This paper introduces Selective Unlearning, a method that unlearns only critical tokens related to unwanted information in large language models, thereby effectively removing specific data while preserving overall utility.

Contribution

The paper proposes a novel Selective Unlearning approach that targets only essential tokens for unlearning, improving efficiency and utility preservation compared to traditional methods.

Findings

Effective unlearning on targeted data

Significant preservation of model utility

Outperforms baseline unlearning algorithms

Abstract

Large Language Model (LLM) unlearning has recently gained significant attention, driven by the need to remove unwanted information, such as private, sensitive, or copyrighted content, from LLMs. However, conventional unlearning approaches indiscriminately update model parameters to forget all tokens in a target document, including common tokens (e.g., pronouns, prepositions, general nouns) that carry general knowledge. In this paper, we highlight that not every token needs forgetting. We propose Selective Unlearning (SU), which identifies a critical subset of tokens within the forgetting set that is relevant to the unwanted information, and unlearns only those tokens. Experiments on two benchmarks and six baseline unlearning algorithms demonstrate that SU not only achieves effective unlearning on the targeted forget data, but also significantly preserves the model's utility in the retaining set.