Get Started
Home
Topics
Search
Library
Research questionHow can visual-document RAG adapt page retrieval to each query without hurting answer accuracy?Fixed page counts can send irrelevant visual-document pages to the vision-language model, increasing generation latency and sometimes weakening its answer. The needed page count may vary with the query’s complexity and relevant content.
Computer Vision
Inference Optimization
Information Retrieval
Multimodal Models
Retrieval-Augmented Generation
Latest papersRecent research connected to this question, newest first.ViSAR: Training-Free Adaptive-$k$ Retrieval for Visual Document Question AnsweringThe source studies training-free adaptive-k retrieval with late-interaction visual document encoders. It constructs a query-conditioned page-level similarity matrix in embedding space and evaluates retrieved page sets across multiple encoders and large vision-language models, reporting up to 58.7% lower RAG latency while maintaining or improving answer accuracy relative to fixed top-k and adaptive heuristics.research paper · Sep 3, 2026
Related questions
How can long-context RAG preserve global document structure during retrieval?How can long-form RAG answer multi-hop questions across changing temporal, spatial, and relational contexts?How can structured RAG synthesize evidence scattered across many documents while limiting answer-generation cost?How can document retrieval find the correct record when large collections share nearly identical visual templates?