Multi-Objective Large Language Model Unlearning

TL;DR

This paper introduces MOLLM, a multi-objective optimization approach for large language model unlearning that effectively removes specific data influence while maintaining overall model utility, addressing gradient explosion and forgetting issues.

Contribution

We propose MOLLM, a novel multi-objective unlearning algorithm for LLMs that improves unlearning effectiveness and utility preservation over existing gradient ascent methods.

Findings

MOLLM outperforms state-of-the-art GA-based methods in unlearning effectiveness.

MOLLM better preserves model utility after unlearning.

The approach effectively mitigates gradient explosion and catastrophic forgetting.

Abstract

Machine unlearning in the domain of large language models (LLMs) has attracted great attention recently, which aims to effectively eliminate undesirable behaviors from LLMs without full retraining from scratch. In this paper, we explore the Gradient Ascent (GA) approach in LLM unlearning, which is a proactive way to decrease the prediction probability of the model on the target data in order to remove their influence. We analyze two challenges that render the process impractical: gradient explosion and catastrophic forgetting. To address these issues, we propose Multi-Objective Large Language Model Unlearning (MOLLM) algorithm. We first formulate LLM unlearning as a multi-objective optimization problem, in which the cross-entropy loss is modified to the unlearning version to overcome the gradient explosion issue. A common descent update direction is then calculated, which enables the model to forget the target data while preserving the utility of the LLM. Our empirical results verify that MoLLM outperforms the SOTA GA-based LLM unlearning methods in terms of unlearning effect and model utility preservation. The source code is available at https://github.com/zibinpan/MOLLM.