Uncovering and Aligning Anomalous Attention Heads to Defend Against NLP Backdoor Attacks
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913028315807744 |
|---|---|
| author | Jin, Haotian Li, Yang Fan, Haihui Shen, Lin Li, Xiangfang Li, Bo |
| author_facet | Jin, Haotian Li, Yang Fan, Haihui Shen, Lin Li, Xiangfang Li, Bo |
| contents | Backdoor attacks pose a serious threat to the security of large language models (LLMs), causing them to exhibit anomalous behavior under specific trigger conditions. The design of backdoor triggers has evolved from fixed triggers to dynamic or implicit triggers. This increased flexibility in trigger design makes it challenging for defenders to identify their specific forms accurately. Most existing backdoor defense methods are limited to specific types of triggers or rely on an additional clean model for support. To address this issue, we propose a backdoor detection method based on attention similarity, enabling backdoor detection without prior knowledge of the trigger. Our study reveals that models subjected to backdoor attacks exhibit unusually high similarity among attention heads when exposed to triggers. Based on this observation, we propose an attention safety alignment approach combined with head-wise fine-tuning to rectify potentially contaminated attention heads, thereby effectively mitigating the impact of backdoor attacks. Extensive experimental results demonstrate that our method significantly reduces the success rate of backdoor attacks while preserving the model's performance on downstream tasks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_13789 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Uncovering and Aligning Anomalous Attention Heads to Defend Against NLP Backdoor Attacks Jin, Haotian Li, Yang Fan, Haihui Shen, Lin Li, Xiangfang Li, Bo Cryptography and Security Artificial Intelligence Backdoor attacks pose a serious threat to the security of large language models (LLMs), causing them to exhibit anomalous behavior under specific trigger conditions. The design of backdoor triggers has evolved from fixed triggers to dynamic or implicit triggers. This increased flexibility in trigger design makes it challenging for defenders to identify their specific forms accurately. Most existing backdoor defense methods are limited to specific types of triggers or rely on an additional clean model for support. To address this issue, we propose a backdoor detection method based on attention similarity, enabling backdoor detection without prior knowledge of the trigger. Our study reveals that models subjected to backdoor attacks exhibit unusually high similarity among attention heads when exposed to triggers. Based on this observation, we propose an attention safety alignment approach combined with head-wise fine-tuning to rectify potentially contaminated attention heads, thereby effectively mitigating the impact of backdoor attacks. Extensive experimental results demonstrate that our method significantly reduces the success rate of backdoor attacks while preserving the model's performance on downstream tasks. |
| title | Uncovering and Aligning Anomalous Attention Heads to Defend Against NLP Backdoor Attacks |
| topic | Cryptography and Security Artificial Intelligence |
| url | https://arxiv.org/abs/2511.13789 |