ScholarPulse 日报 2026-11-11

2026-11-11 学术简报：2 篇。针对大模型科学智能体否定单一假设后缺乏备选方向的问题，本文提出“对比发现”范式，通过评估互斥且完备的竞争解释集，在相同实验预算下显著提升了科学问题的解决率。

今日速览

序号	标题	来源	日期	主题	推荐等级
1	Contrastive Discovery: Open-Ended Scientific Discovery over Competing Explanations	arXiv	2026-11-10	AI-Agent	高
2	ReTrace: Distilling Tool-Using Agents into Planner-Free Workflows	arXiv	2026-11-10	AI-Agent	高

重点论文与技术动态

1. Contrastive Discovery: Open-Ended Scientific Discovery over Competing Explanations

来源：arXiv
日期：2026-11-10
作者/机构：Ziang Liu, James J. Kim, Yijia Dai, Jennifer J. Sun
主题标签：AI-Agent, arXiv
推荐等级：高
分类：cs.AI, cs.CL, cs.LG

一句话结论

针对大模型科学智能体否定单一假设后缺乏备选方向的问题，本文提出“对比发现”范式，通过评估互斥且完备的竞争解释集，在相同实验预算下显著提升了科学问题的解决率。

核心内容

逐点评估局限：跨6个领域中，智能体发生大幅信念变化的候选假设有89%仅为否定，无法得知证据究竟支持何种替代解释。
对比发现机制：将单一假设替换为互斥且完备的“对比集”，依据证据对信念的重新分配，当信念集中于某一解释时即解决问题。
综合性能突破：在相同证据下解决了73%被逐点拒绝的问题；引导搜索时比逐点方法多解决67%的问题，与38位专家的盲测一致性达78%~82%。

方法与数据

将“对比集”作为提议与评分单元；在6个领域任务及38名人类专家的双盲实验中完成测试与验证。

价值判断

值得关注：证明了显式引入竞争性解释能在相同证据与验证预算下，大幅增强智能体解决科学问题的能力。
可复用点：对比集重分配信念的评分机制可直接用于现有智能体，重估已有数据或引导新的假设搜索。
局限/待核查：如何针对开放场景构建严格互斥且完备的解释集，摘要未明确。

摘要

LLM agents are increasingly used for open-ended, data-driven scientific discovery, yet most of them evaluate hypotheses pointwise, testing an empirical claim $h$ only against its complement $\neg h$. When evidence rejects $h$, the agent learns that one explanation is unlikely but not which alternative the evidence supports. To recover that alternative, we introduce Contrastive Discovery, which replaces the single hypothesis with a contrast set, a scientific question paired with mutually exclusive and collectively exhaustive competing explanations. Each experiment is scored by how its evidence redistributes belief across the set, and the question is resolved when belief concentrates on one explanation. Because contrast sets change only the unit an agent proposes and scores, an existing agent can use them to re-evaluate evidence it has already collected or to guide its search. Across six domains, $89\%$ of the pointwise candidates with the largest belief shifts are rejections. On the same evidence, contrast sets resolve $73\%$ of these candidates to a specific alternative. When contrast sets guide search, the agent resolves $67\%$ more questions than pointwise surprisal under the same experiment budget. In a blinded study, $38$ human experts chose the same explanation on $78\%$ of resolved questions and on $82\%$ of questions where the pointwise agent rejected its hypothesis and stopped. Together, these results support explicit competing explanations as a way for discovery agents to resolve more scientific questions from the same evidence and validation budget.

2. ReTrace: Distilling Tool-Using Agents into Planner-Free Workflows

来源：arXiv
日期：2026-11-10
作者/机构：Kaiyuan Hou, Yang Ping Kuo, Indranil Bhattacharya, Rohith Nama, Arushi Gupta, Carlos D. Castillo
主题标签：AI-Agent, arXiv
推荐等级：高
分类：cs.AI, cs.LG, cs.MA

一句话结论

ReTrace 通过将 SOP 和少量 Agent 运行日志蒸馏为由代码控制流驱动的可执行工作流，在保持与 Agent 相当成功率的同时大幅降低了延迟与成本。

核心内容

针对 Agent 逐步调用 LLM 规划导致的高延迟与成本问题，改由代码替代 LLM 选择下一步动作。
利用 Agent 调用工具时的推理说明提取步骤条件，合并为紧凑规则文档，再生成目标平台的执行代码。
引入失败修复机制，在大幅缩短运行时间（如从 170s 降至 1.3s）的同时，成功率可进一步超越原 Agent（如 87.8% 对比 85.5%）。

方法与数据

包含条件提取、跨运行合规化规则及平台工作流编译；评估使用了 79 个生产流程、 $\tau^2$ -Bench 以及 8 个 SOP-Bench 流程。

价值判断

值得关注：实现了百倍级的推理延迟缩减与成本节约（如生产任务成本降 24%），显著提升 SOP 自动化效率。
可复用点：规则文档与执行平台解耦，同一份逻辑规则可编译适配不同的流程编排平台。
局限/待核查：未经修复时初始表现略逊于 Agent，达到高成功率依赖基于带标签训练失败样本的修复轮次。

摘要

Business process automation runs the same Standard Operating Procedure (SOP) again and again on new inputs. Large Language Model (LLM) agents can follow an SOP written in natural language, but they call an LLM to choose every step of every run, which adds latency and cost at scale. We present ReTrace, which turns an SOP and a few logged agent runs into a reusable workflow in which code, not an LLM, selects the next step. The agent's reasoning, the explanation it writes with each tool call, often states why it took an action; the tool calls alone do not. ReTrace extracts the condition behind each step, merges these conditions across runs into a compact rule document, and generates an executable workflow from it. Only this last stage depends on the platform, so rule documents can be compiled for different orchestration platforms. On $79$ procedures from our production platform, ReTrace workflows reach $93.8\%$ completion, against $96.8\%$ for the agent, at $24\%$ lower cost per task. On $τ^2$-Bench, they reach $75.8\%$ task success (agent: $80.6\%$) and run in $1.3$s instead of $170$s; three rounds of repair on training failures lift $k{=}2$ workflows from $66.7\%$ to $81.2\%$. On eight SOP-Bench procedures, ReTrace without repair trails few-shot generation ($80.6\%$ vs. $83.6\%$), but repair from labeled training failures lifts it to $87.8\%$ at $0.40$s per task, compared to $85.5\%$ in $46.6$s for the agent.