Cut the Deadwood Out: Backdoor Purification via Guided Module Substitution

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Tong, Yao, Li, Weijun, He, Xuanli, Zhan, Haolan, Xu, Qiongkai
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911184323608576
author Tong, Yao
Li, Weijun
He, Xuanli
Zhan, Haolan
Xu, Qiongkai
author_facet Tong, Yao
Li, Weijun
He, Xuanli
Zhan, Haolan
Xu, Qiongkai
contents Model NLP models are commonly trained (or fine-tuned) on datasets from untrusted platforms like HuggingFace, posing significant risks of data poisoning attacks. A practical yet underexplored challenge arises when such backdoors are discovered after model deployment, making retraining-required defenses less desirable due to computational costs and data constraints. In this work, we propose Guided Module Substitution (GMS), an effective retraining-free method based on guided merging of the victim model with just a single proxy model. Unlike prior ad-hoc merging defenses, GMS uses a guided trade-off signal between utility and backdoor to selectively replaces modules in the victim model. GMS offers four desirable properties: (1) robustness to the choice and trustworthiness of the proxy model, (2) applicability under inaccurate data knowledge, (3) stability across hyperparameters, and (4) transferability across different attacks. Extensive experiments on encoder models and decoder LLMs demonstrate the strong effectiveness of GMS. GMS significantly outperforms even the strongest defense baseline, particularly against challenging attacks like LWS.
format Preprint
id arxiv_https___arxiv_org_abs_2412_20476
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Cut the Deadwood Out: Backdoor Purification via Guided Module Substitution
Tong, Yao
Li, Weijun
He, Xuanli
Zhan, Haolan
Xu, Qiongkai
Computation and Language
Cryptography and Security
Model NLP models are commonly trained (or fine-tuned) on datasets from untrusted platforms like HuggingFace, posing significant risks of data poisoning attacks. A practical yet underexplored challenge arises when such backdoors are discovered after model deployment, making retraining-required defenses less desirable due to computational costs and data constraints. In this work, we propose Guided Module Substitution (GMS), an effective retraining-free method based on guided merging of the victim model with just a single proxy model. Unlike prior ad-hoc merging defenses, GMS uses a guided trade-off signal between utility and backdoor to selectively replaces modules in the victim model. GMS offers four desirable properties: (1) robustness to the choice and trustworthiness of the proxy model, (2) applicability under inaccurate data knowledge, (3) stability across hyperparameters, and (4) transferability across different attacks. Extensive experiments on encoder models and decoder LLMs demonstrate the strong effectiveness of GMS. GMS significantly outperforms even the strongest defense baseline, particularly against challenging attacks like LWS.
title Cut the Deadwood Out: Backdoor Purification via Guided Module Substitution
topic Computation and Language
Cryptography and Security
url https://arxiv.org/abs/2412.20476