Recursive Agentic Reasoning

📄 arXiv: 2608.23956v1 📥 PDF

作者: Shengxin Zhang, Xiaomin Wu, Xiyang Wu, Jing Xie

分类: cs.AI

发布日期: 2026-08-25


💡 一句话要点

提出统一视角的递归推理方法以提升模型性能

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

关键词: 递归推理 模型评估 自然语言处理 智能问答 决策支持

📋 核心要点

  1. 现有的测试时推理方法往往孤立评估,导致不同模型和基准之间的比较困难。
  2. 本文提出了将推理方法视为递归算子,包括GROW、PRUNE和BRANCH,以统一评估这些方法的效果。
  3. 实验结果表明,BRANCH在14个设置中平均提高了5.98个百分点,是表现最佳的算子,显示出重复分支的优势。

📝 摘要(中文)

现有的测试时推理方法如迭代精炼、分解和重复采样常常孤立评估,导致不同模型、基准和评估流程之间的比较困难。本文提出了将这些方法视为代理推理轨迹上的递归算子:GROW、PRUNE和BRANCH。我们在相同的评估环境下对这三种算子进行评估,结果显示BRANCH在所有14个设置中平均提高了5.98个百分点,并在12个设置中表现最佳。相较之下,GROW的平均增益为2.18个百分点,并在两个设置中表现下降,而PRUNE的平均提升为0.94个百分点。分析表明,BRANCH的优势不仅在于探索多个推理路径,还在于从截断中恢复。最后,我们强调了成对评分的重要性,以避免评估结果的偏差。

🔬 方法详解

问题定义:本文旨在解决现有测试时推理方法孤立评估的问题,导致不同模型和基准之间的比较困难。现有方法缺乏统一的视角,难以评估其相对性能。

核心思路:论文提出将推理方法视为递归算子,包括GROW(深化推理路径)、PRUNE(分解和重组问题)和BRANCH(采样替代推理路径),以便在相同的评估环境下进行比较。

技术框架:整体架构包括三个主要模块:GROW、PRUNE和BRANCH。每个模块在相同的提示、令牌预算和评分代码下进行评估,确保公平比较。

关键创新:最重要的技术创新在于将不同的推理方法统一视为递归算子,并通过实验验证了BRANCH在多种设置中的优越性,挑战了不同问题需要不同推理算子的假设。

关键设计:在实验中,使用了相同的提示和评分标准,确保了评估的一致性。BRANCH的优势与基线模型输出的空白率(r = 0.72)高度相关,表明其在处理截断时的恢复能力。

🖼️ 关键图片

img_0
img_1
img_2

📊 实验亮点

实验结果显示,BRANCH在14个设置中平均提高了5.98个百分点,并在12个设置中表现最佳。相比之下,GROW的平均增益为2.18个百分点,PRUNE的平均提升为0.94个百分点,显示出BRANCH的显著优势。

🎯 应用场景

该研究的潜在应用领域包括自然语言处理、智能问答系统和决策支持系统等。通过提升推理能力,模型能够在复杂任务中表现更佳,具有重要的实际价值和未来影响。

📄 摘要(原文)

Test-time reasoning methods such as iterative refinement, decomposition, and repeated sampling are often evaluated in isolation, making their gains difficult to compare across models, benchmarks, and evaluation pipelines. We introduce a unified view of these methods as recursion operators over an agent's reasoning trace: GROW, which deepens a single reasoning path; PRUNE, which decomposes and recomposes the problem; and BRANCH, which samples alternative reasoning paths and selects among them. We evaluate all three operators against a single-pass chain-of-thought baseline under a shared harness with identical prompts, token budgets, and grading code. Across five benchmarks and three frontier models, comprising 14 model-benchmark settings, 49,327 graded items, and 151,876 model calls, BRANCH improves accuracy in all 14 settings by an average of 5.98 percentage points and is the best-performing operator in 12. In contrast, GROW yields a mean gain of 2.18 points and degrades performance in two settings, while PRUNE improves accuracy by 0.94 points on average. Analysis shows that BRANCH's advantage arises not only from exploring multiple reasoning paths, but also from recovering from truncation: its gains strongly correlate with the baseline rate of empty, budget-exhausted outputs (r = 0.72). These results weaken the hypothesis that different problems require routing among test-time reasoning operators; at this level of abstraction, repeated branching is consistently dominant. Finally, we show that unpaired evaluation and treating scoring-pipeline failures as model errors can materially change, and even reverse, comparative conclusions, motivating paired scoring as a standard protocol for test-time-compute evaluation.