AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Uncovering and Aligning Anomalous Attention Heads to Defend Against NLP Backdoor Attacks

Haotian Jin, Yang Li, Haihui Fan, Lin Shen, Xiangfang Li, Bo Li

AAAI 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Trigger-based backdoor attacks · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add attention-head alignment to head-wise fine-tuning (ablation)Fine-tuning-only defense · SST-2 BadNets, BERT-baseASR: 53.37 -> 11.23Improved
add attention-head alignment to head-wise fine-tuning (ablation)Fine-tuning-only defense · SST-2 BadNets, Llama2-7BASR: 56.17 -> 25.61Improved
add attention-head alignment to head-wise fine-tuning (ablation)Fine-tuning-only defense · Targeted refusal Sleeper Agent, Mistral-7BASR: 31.09 -> 8.54Improved
add head-wise fine-tuning after alignment (ablation)Alignment-only defense · SST-2 BadNets, BERT-baseclean accuracy: 83.39 -> 91.09Improved
add head-wise fine-tuning after alignment (ablation)Alignment-only defense · SST-2 BadNets, Llama2-7Bclean accuracy: 80.15 -> 85.19Improved
add head-wise fine-tuning after alignment (ablation)Alignment-only defense · SST-2 BadNets, Llama2-7BASR: 23.41 -> 25.61No clear effect
add head-wise fine-tuning after alignment (ablation)Alignment-only defense · SST-2 BadNets, BERT-baseASR: 18.36 -> 11.23Improved
use proposed defense instead of PUREBERT-base · SST-2 HiddenKillerASR: 35.26 -> 15.66Improved
use proposed defense instead of GraCeFulLlama2-7B · Sentiment steering VPIASR: 10.98 -> 24.57Hurt
use proposed defense instead of MuScleLoRALlama2-7B · Sentiment steering VPIclean accuracy: 78.19 -> 87.13Improved
vary safety threshold tau (ablation)Attention head classification · backdoor defenseASR/CANo clear effect
use learning rate 2e-4 for safe heads and 5e-6 for suspicious heads (ablation)Head-wise fine-tuning · backdoor defenseASR and CAImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy