Backdoor defense for LLMs detects anomalous attention-head similarity on triggers and applies head-wise alignment via fine-tuning to reduce attack success.
The final PCA-reduced dimensionalityh i is set to 32
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Uncovering and Aligning Anomalous Attention Heads to Defend Against NLP Backdoor Attacks
Backdoor defense for LLMs detects anomalous attention-head similarity on triggers and applies head-wise alignment via fine-tuning to reduce attack success.