Reproduce “Safeguarding Multimodal Knowledge Copyright in the RAG-as-a-Service Environment” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tianyu Chen, Jian Lou, Wenjie Wang
AQUA tests whether an unauthorized RAG service uses a provider’s images by embedding semantic clues in synthetic images and probing for their expected textual signatures. It evaluates acronym and spatial-relation methods on MMQA and WebQA with four vision-language generators. For Qwen2.5-VL-Instruct (7B) on MMQA, acronym CGSR is 99.61%, conditional on image retrieval, versus 19.96% for Opt.; findings apply to tested setups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.