OpenLVLM-MIA: A Controlled Benchmark Revealing the Limits of Membership Inference Attacks on Large Vision-Language Models
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866918227114721280 |
|---|---|
| author | Miyamoto, Ryoto Fan, Xin Kido, Fuyuko Matsumoto, Tsuneo Yamana, Hayato |
| author_facet | Miyamoto, Ryoto Fan, Xin Kido, Fuyuko Matsumoto, Tsuneo Yamana, Hayato |
| contents | OpenLVLM-MIA is a new benchmark that highlights fundamental challenges in evaluating membership inference attacks (MIA) against large vision-language models (LVLMs). While prior work has reported high attack success rates, our analysis suggests that these results often arise from detecting distributional bias introduced during dataset construction rather than from identifying true membership status. To address this issue, we introduce a controlled benchmark of 6{,}000 images where the distributions of member and non-member samples are carefully balanced, and ground-truth membership labels are provided across three distinct training stages. Experiments using OpenLVLM-MIA demonstrated that the performance of state-of-the-art MIA methods approached chance-level. OpenLVLM-MIA, designed to be transparent and unbiased benchmark, clarifies certain limitations of MIA research on LVLMs and provides a solid foundation for developing stronger privacy-preserving techniques. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_16295 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | OpenLVLM-MIA: A Controlled Benchmark Revealing the Limits of Membership Inference Attacks on Large Vision-Language Models Miyamoto, Ryoto Fan, Xin Kido, Fuyuko Matsumoto, Tsuneo Yamana, Hayato Computer Vision and Pattern Recognition Artificial Intelligence OpenLVLM-MIA is a new benchmark that highlights fundamental challenges in evaluating membership inference attacks (MIA) against large vision-language models (LVLMs). While prior work has reported high attack success rates, our analysis suggests that these results often arise from detecting distributional bias introduced during dataset construction rather than from identifying true membership status. To address this issue, we introduce a controlled benchmark of 6{,}000 images where the distributions of member and non-member samples are carefully balanced, and ground-truth membership labels are provided across three distinct training stages. Experiments using OpenLVLM-MIA demonstrated that the performance of state-of-the-art MIA methods approached chance-level. OpenLVLM-MIA, designed to be transparent and unbiased benchmark, clarifies certain limitations of MIA research on LVLMs and provides a solid foundation for developing stronger privacy-preserving techniques. |
| title | OpenLVLM-MIA: A Controlled Benchmark Revealing the Limits of Membership Inference Attacks on Large Vision-Language Models |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2510.16295 |