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

Open in new window