Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Fan, Chongyu, Liu, Jiancheng, Lin, Licong, Jia, Jinghan, Zhang, Ruiqi, Mei, Song, Liu, Sijia
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914100205846528
author Fan, Chongyu
Liu, Jiancheng
Lin, Licong
Jia, Jinghan
Zhang, Ruiqi
Mei, Song
Liu, Sijia
author_facet Fan, Chongyu
Liu, Jiancheng
Lin, Licong
Jia, Jinghan
Zhang, Ruiqi
Mei, Song
Liu, Sijia
contents This work studies the problem of large language model (LLM) unlearning, aiming to remove unwanted data influences (e.g., copyrighted or harmful content) while preserving model utility. Despite the increasing demand for unlearning, a technically-grounded optimization framework is lacking. Gradient ascent (GA)-type methods, though widely used, are suboptimal as they reverse the learning process without controlling optimization divergence (i.e., deviation from the pre-trained state), leading to risks of over-forgetting and potential model collapse. Negative preference optimization (NPO) has been proposed to address this issue and is considered one of the state-of-the-art LLM unlearning approaches. In this work, we revisit NPO and identify another critical issue: reference model bias. This bias arises from using the reference model (i.e., the model prior to unlearning) to evaluate the unlearning success, which can compromise NPO's effectiveness. Specifically, it leads to (a) uneven allocation of optimization power across forget data with varying difficulty levels and (b) ineffective gradient weight smoothing during the early stages of unlearning optimization. To overcome these challenges, we propose a simple yet effective unlearning optimization framework, called SimNPO, showing that `simplicity' in removing the reliance on a reference model (through the lens of simple preference optimization) benefits unlearning. We provide deeper insights into SimNPO's advantages through an analysis based on mixtures of Markov chains. Extensive experiments further validate SimNPO's efficacy on benchmarks like TOFU and MUSE, as well as its robustness against relearning attacks. Codes are available at https://github.com/OPTML-Group/Unlearn-Simple.
format Preprint
id arxiv_https___arxiv_org_abs_2410_07163
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning
Fan, Chongyu
Liu, Jiancheng
Lin, Licong
Jia, Jinghan
Zhang, Ruiqi
Mei, Song
Liu, Sijia
Computation and Language
Artificial Intelligence
Machine Learning
This work studies the problem of large language model (LLM) unlearning, aiming to remove unwanted data influences (e.g., copyrighted or harmful content) while preserving model utility. Despite the increasing demand for unlearning, a technically-grounded optimization framework is lacking. Gradient ascent (GA)-type methods, though widely used, are suboptimal as they reverse the learning process without controlling optimization divergence (i.e., deviation from the pre-trained state), leading to risks of over-forgetting and potential model collapse. Negative preference optimization (NPO) has been proposed to address this issue and is considered one of the state-of-the-art LLM unlearning approaches. In this work, we revisit NPO and identify another critical issue: reference model bias. This bias arises from using the reference model (i.e., the model prior to unlearning) to evaluate the unlearning success, which can compromise NPO's effectiveness. Specifically, it leads to (a) uneven allocation of optimization power across forget data with varying difficulty levels and (b) ineffective gradient weight smoothing during the early stages of unlearning optimization. To overcome these challenges, we propose a simple yet effective unlearning optimization framework, called SimNPO, showing that `simplicity' in removing the reliance on a reference model (through the lens of simple preference optimization) benefits unlearning. We provide deeper insights into SimNPO's advantages through an analysis based on mixtures of Markov chains. Extensive experiments further validate SimNPO's efficacy on benchmarks like TOFU and MUSE, as well as its robustness against relearning attacks. Codes are available at https://github.com/OPTML-Group/Unlearn-Simple.
title Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2410.07163