今日速览
| 序号 | 标题 | 来源 | 日期 | 主题 | 推荐等级 |
|---|---|---|---|---|---|
| 1 | Navigating Sparse Evidence: Agentic Visual RAG via Explicit Context Selection and Consolidation | arXiv | 2026-09-14 | RAG | 高 |
| 2 | Question’s Gambit: The First Move Matters in Agentic Deep Search | arXiv | 2026-09-13 | RAG | 高 |
重点论文与技术动态
1. Navigating Sparse Evidence: Agentic Visual RAG via Explicit Context Selection and Consolidation
- 来源:arXiv
- 日期:2026-09-14
- 作者/机构:Yucheng Shen, Lingyong Yan, Jiulong Wu, Shuaiqiang Wang, Jianmin WU, Dawei Yin
- 主题标签:
RAG,arXiv - 推荐等级:高
- 分类:cs.AI
一句话结论
SCoRE 通过显式的证据选择与整合机制,将视觉 RAG 中的探索过程与最终推理解耦,缓解了稀疏视觉证据利用不足的问题。
核心内容
- 现有 VRAG 方法依赖原始探索轨迹或压缩文本记忆生成答案,易受探索噪声干扰且证据链不透明。
- SCoRE 在探索阶段仅保留查询相关的观察及来源指针于文本账本中,终止时重新加载原始图像并整合为逻辑序列,实现严格的视觉 grounding。
- 训练范式结合过滤冷启动轨迹蒸馏与证据感知强化学习,奖励函数同时优化证据覆盖、整合紧凑性和答案正确性。
方法与数据
- 方法:统一 agent 循环(选择-整合),配合冷启动蒸馏 + 证据感知 RL 的端到端训练。
- 数据:摘要未明确。
价值判断
- 值得关注:将”证据组织”作为独立瓶颈提出,思路清晰,对多模态 RAG 系统设计有启发。
- 可复用点:文本账本 + 来源指针的轻量证据管理策略,以及证据感知奖励设计可迁移至其他 agentic 场景。
- 局限/待核查:摘要未给出具体基准和对比实验结果,实际效果需查阅正文验证。
摘要
Visual Retrieval-Augmented Generation (VRAG) empowers models to navigate and answer queries about visually rich documents by retrieving relevant page images as visual evidence and reasoning over their content. However, effectively utilizing this visual evidence is usually impeded by two main challenges. First, answer-relevant evidence is sparse and may be concentrated in a small region of one page or dispersed across multiple pages. Second, existing agentic methods often generate answers based on raw exploration trajectories or compressed textual memories rather than an explicitly organized set of supporting images, making answers susceptible to exploration noise and obscuring the evidence-backed reasoning trace. We argue that the bottleneck lies not only in evidence discovery but also in its preservation and organization before answer generation. We propose SCoRE (Selection and Consolidation for Robust Evidence), a unified agent loop for explicit evidence selection and consolidation. During exploration, SCoRE retains only query-relevant observations and their source pointers in a maintained textual ledger, preserving earlier evidence while keeping the visual context bounded. At termination, it reloads the referenced original images and consolidates the visual evidence for answering, arranging it into a logical sequence. This decouples final reasoning from exploratory trial-and-error while ensuring strict visual grounding via indexed claim-to-image linkages. To enable end-to-end optimization of this unified rollout, our training paradigm combines filtered cold-start trajectory distillation with evidence-aware reinforcement learning, whose reward promotes evidence coverage, consolidation compactness, and answer correctness.2. Question’s Gambit: The First Move Matters in Agentic Deep Search
- 来源:arXiv
- 日期:2026-09-13
- 作者/机构:Radin Hamidi Rad, Amin Bigdeli, Negar Arabzadeh, Sajad Ebrahimi, Charles L. A. Clarke, Benjamin C. M. Fung
- 主题标签:
RAG,arXiv - 推荐等级:高
- 分类:cs.AI, cs.CL, cs.IR
一句话结论
深度研究代理的首次检索动作是决定最终表现的关键设计决策,Question’s Gambit 通过优化首步检索将答案准确率从 83.1% 提升至 90.5%。
核心内容
- 深度研究代理在迭代搜索-推理循环中,即使检索配置良好,仍可能无法将证据文档与黄金文档有效关联。
- Question’s Gambit 将问题分解为线索集,重构为互补搜索,整合并重排序候选池,为代理构建支持线索聚合与答案验证的初始上下文。
- 在 BrowseComp-Plus 上,使用 gpt-5.5 时答案准确率从 83.1% 提升至 90.5%,超越最强基线 Pi-Serini。
方法与数据
- 方法:首步检索模块(线索分解→互补搜索→结果整合→重排序),在代理迭代循环之前执行。
- 数据:BrowseComp-Plus(主实验)、MultiHop-RAG(迁移验证)。
价值判断
- 值得关注:首次检索作为独立设计维度的提出,为 agentic deep search 提供了新的优化切入点。
- 可复用点:线索分解+互补搜索+重排序的模块化流程可直接嵌入现有代理框架。
- 局限/待核查:MultiHop-RAG 上的具体提升幅度摘要未明确;gpt-5.5 的可用性需确认。