AI RESEARCH
EntropyScan: Towards Model-level Backdoor Detection in LVLMs via Visual Attention Entropy
arXiv CS.CV
•
ArXi:2605.15711v1 Announce Type: new Large Vision-Language Models (LVLMs) have nstrated remarkable capabilities across various tasks, yet they remain vulnerable to backdoor attacks. Existing defense methods predominantly focus on sample-level defense, which relies on the knowledge of