COMPASS: Membership Inference Attack Against Vision Large Language Models via Adaptive Noise Calibration

Summary

COMPASS is an adaptive gray-box membership inference attack against vision large language models (VLLMs) that calibrates both perturbation magnitude and membership-score direction using a known non-member reference set. The authors find that a low-noise regime often separates members from non-members better than large perturbations, with the score direction reversing in some model-dataset pairs. Tested on LLaVA v1.5-7B, MiniGPT-4 and HuluMed-32B, COMPASS reaches AUCs up to 0.88 on general-domain data and 0.64 on distribution-matched medical data, where the strongest baseline stays at chance.

Topics

• Training data and privacy

Related items

• InfoA novel privacy-preserving large language model integrating trust-weighted and ethical gradient maskingSimilar attack · OpenAlex (peer-reviewed AI security)
• InfoKinematics-Induced Multimodal 3D Human Pose Estimation with Subject-Level PrivacySimilar attack · Arxiv (cs.RO + cs.CV security)
• InfoGDPFDL: A secure and efficient personalized federated distillation framework against inference attacksSimilar attack · Elsevier Security Journals
• InfoBenchmarking post-processing methods in local differential privacy for utility and adversarial robustnessSimilar attack · Elsevier Security Journals
• InfoDeepU: Deeper Granular Within-Layer Machine UnlearningSimilar attack · IEEE Xplore (Security & AI Journals)