CLIMB: Confidence-Guided Complementary Evidence for Multimodal Retrieval-Augmented Generation

类型：arxiv
标识：2610.03421
链接：http://arxiv.org/abs/2610.03421v1
主分类：rag
形态：method
TLDR：Multimodal large language models (MLLMs) have shown strong visual reasoning abilities, but knowledge-intensive visual question answering often requires external textual evidence beyond the image and the model's parametric knowledge. Existing multimodal RAG systems commonly rely on Top-$K$ retrieval or reranking, which may return redundant passages and provide limited control over whether an answer update is sufficiently supported by the retrieved evidence. We propose \textit{CLIMB}, a training-free inference-time framework for multimodal RAG. CLIMB first constructs a compact complementary evid
副分类：multimodal
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-10-05-agent-rag-longcontext-candidates.json

2026-09-14

笔记 MMLongEmbed: 多模态嵌入模型长上下文基准测试

2026-06-17

笔记 flyP 精读与批判 · 2026-07-29 22:50（v2 覆盖）

2026-07-29

笔记 BRIDGE:长多模态文档多跳推理基准(短审稿 · flyP)

2026-07-03

笔记 2026-07-04 上午多模态对照精读 · Seeker（文本→像素的长上下文 MLLM）

2026-07-04

笔记 2026-09-05 精读批判 | 多模态长文档 RAG 与高效 VLM · v2 覆盖版

2026-09-05

arXiv

2610.03421

TLDR

Multimodal large language models (MLLMs) have shown strong visual reasoning abilities, but knowledge-intensive visual question answering often requires external textual evidence beyond the image and the model's parametric knowledge. Existing multimodal RAG systems commonly rely on Top-$K$ retrieval or reranking, which may return redundant passages and provide limited control over whether an answer update is sufficiently supported by the retrieved evidence. We propose \textit{CLIMB}, a training-free inference-time framework for multimodal RAG. CLIMB first constructs a compact complementary evid

CLIMB: Confidence-Guided Complementary Evidence for Multimodal Retrieval-Augmented Generation

相关 · 按语义匹配