Adversary-Aware DPO: Enhancing Safety Alignment in Vision Language Models via Adversarial Training

AAML
Re-assign community