Image-to-Text Membership Inference on COCO 2017
91ASROurs
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OursBackbone Model=LLaVA-7B, Knowledge Setting=Partial Knowledge2026.07 | 91 | 97 | 55 | |
| MaxRenyi-K% MIABackbone Model=LLaVA-7B, Knowledge Setting=Partial Knowledge2026.07 | 89 | 94 | 52 | |
| unified membership inference frameworkKnowledge Setting=Zero-Knowledge, Base Model=LLaVA-7B, Embedding Extractor=all-MiniLM-L6-v22026.07 | 88 | 96 | 53 | |
| Temperature-based MIABackbone Model=LLaVA-7B, Knowledge Setting=Partial Knowledge2026.07 | 88 | 95 | 53 | |
| Temp.Knowledge Setting=Zero-Knowledge, Base Model=LLaVA-7B, Embedding Extractor=all-MiniLM-L6-v22026.07 | 60 | 58 | 23 | |
| MaxRenyiKnowledge Setting=Zero-Knowledge, Base Model=LLaVA-7B, Embedding Extractor=all-MiniLM-L6-v22026.07 | 58 | 55 | 19 |