RECAST: Learning to Compute the Right Context through Adaptive Evidence Routing

📄 arXiv: 2610.10507v1 📥 PDF

作者: Yilun Hao, Krishna Sayana, Isabella Ye, James S Ren, Sukhdeep Sodhi, Craig Boutilier, Chuchu Fan

分类: cs.AI

发布日期: 2026-10-07

备注: 35 pages, 2 figures, 13 tables


💡 一句话要点

提出RECAST以解决多源信息证据构建问题

🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)

关键词: 证据构建 异构信息 检索增强生成 顺序决策 机器学习 自然语言处理 智能助手

📋 核心要点

  1. 现有的检索增强生成方法在处理多源信息时,往往无法有效推导所需证据,导致性能受限。
  2. RECAST通过将证据构建视为一个顺序决策过程,结合异构检索和计算操作,主动推导证据。
  3. 在六个基准测试中,RECAST的平均成功率为75.6%,显著超越了现有最强基线,展示了强大的零-shot 泛化能力。

📝 摘要(中文)

随着大型语言模型在长篇异构信息源任务中的应用日益增多,传统的检索增强生成(RAG)方法依赖于固定的相似性检索,而代理变体虽然能够适应查询和工具使用,但仍然以检索为中心。在许多任务中,解决方案所需的证据并不明确存在于任何单一源项中,而是需要通过过滤、聚合或计算从多个源项中推导而来。本文提出了RECAST(通过计算、访问和合成工具路由证据),这是一个将证据构建形式化为异构检索和计算操作的顺序决策过程的学习框架,允许主动推导证据而不仅仅是检索。实验结果显示,RECAST在六个异构基准测试中取得了75.6%的平均成功率,超越了最强大型模型基线15.9%。

🔬 方法详解

问题定义:本文旨在解决在多源信息中有效构建证据的问题。现有的检索增强生成方法往往依赖于固定的相似性检索,无法处理需要从多个源项中推导的证据。

核心思路:RECAST的核心思路是将证据构建视为一个顺序决策过程,允许通过异构检索和计算操作主动推导证据,而不仅仅是被动检索。

技术框架:RECAST的整体架构包括三个主要模块:RouterLM、CompilerLM和AnswerLM。RouterLM负责选择和制定原始操作,CompilerLM将其转化为可执行代码,AnswerLM生成最终答案。

关键创新:RECAST的最大创新在于将证据构建过程形式化为决策过程,允许动态选择操作,从而提高了对复杂任务的适应性和灵活性。

关键设计:RouterLM通过监督微调(SFT)和组相对策略优化(GRPO)进行训练,设计了轻量级的操作选择机制,确保了在多种任务和异构源表示下的有效性。

🖼️ 关键图片

fig_0
fig_1

📊 实验亮点

RECAST在六个异构基准测试中实现了75.6%的平均成功率,超越了最强大型模型基线15.9%。此外,Qwen3.5-9B RouterLM在训练后比无训练的Gemini 3.5 Flash RouterLM提高了5.0%。在三个保留基准上,RECAST平均提升了15.0%,展示了其强大的零-shot 泛化能力。

🎯 应用场景

RECAST的研究成果在多个领域具有广泛的应用潜力,尤其是在需要从多个异构信息源中提取和整合证据的任务中,如法律文书分析、医学诊断支持和复杂问题求解等。其灵活的证据构建能力将推动智能助手和决策支持系统的发展,提升其在实际应用中的表现和可靠性。

📄 摘要(原文)

Large language models are increasingly applied to tasks grounded in long, heterogeneous information sources. Conventional Retrieval-Augmented Generation (RAG) relies on fixed similarity-based retrieval, while agentic variants adapt queries and tool use but remain largely retrieval-centric. However, in many tasks, the evidence required for a solution is not explicitly present in any single source item. Instead, it must be derived through filtering, aggregation, or computation across multiple source items. In this work, we introduce RECAST (Routing Evidence through Computation, Access, and Synthesized Tools), a learned framework that formulates evidence construction as a sequential decision process over heterogeneous retrieval and computation operations, allowing evidence to be actively derived rather than merely retrieved. A lightweight RouterLM iteratively selects and formulates primitive operations or specifies customized operations for a frozen CompilerLM to translate into executable code. Once it judges the evidence sufficient, RouterLM passes the accepted evidence to a frozen AnswerLM to produce the final solution. We train RouterLM with supervised fine-tuning (SFT) followed by group relative policy optimization (GRPO). Across six heterogeneous benchmark families, RECAST achieves a mean success rate of 75.6%, outperforming the strongest large-model baseline by 15.9%. Moreover, training enables the Qwen3.5-9B RouterLM to outperform a training-free Gemini 3.5 Flash RouterLM by 5.0%. On three held-out benchmarks, RECAST improves over the strongest baseline by 15.0% on average, demonstrating strong zero-shot generalization across tasks and heterogeneous source representations.