[论文] Simulate, Reason, Decide: Scientific Reasoning with LLMs for Simulatio...

论文概要

研究领域: ML
作者: Yuhan Yang, Ruipu Li, Alexander Rodríguez
发布时间: 2025-06-01
arXiv: 2606.04505

中文摘要

科学模拟器正日益被整合到LLM驱动的系统中，用于高风险模拟驱动决策。然而，现有框架主要使用LLM来生成、校准或执行模拟器，将其视为黑盒接口，而非可被推理的结构化机制系统。因此，当前方法缺乏识别、表征和推理模拟器行为背后假设和机制的能力，限制了透明度、可审计性和决策论证。我们引入MechSim，一种面向可执行科学模拟器的机制 grounded 神经符号推理框架。与先前主要对静态符号结构进行推理的神经符号方法不同，MechSim使LLM智能体能够对科学模拟器的机制、假设和执行行为进行推理。我们的框架通过共享结构化模式表征模拟器，捕获假设、变量、机制依赖和执行轨迹。在此表征之上，LLM智能体作为约束推理引擎，生成结构化、证据 grounded 的解释，将模拟器结果与其底层机制联系起来。我们在多个高风险领域评估了我们的方法，显示其提升了机制级解释质量、模拟器分析和下游决策可靠性。

原文摘要

Scientific simulators are increasingly being integrated into LLM-driven systems for high-stakes simulation-driven decision-making. However, existing frameworks primarily use LLMs to generate, calibrate, or execute simulators, treating them as black-box interfaces rather than as structured mechanistic systems that can be reasoned about. As a result, current approaches lack the ability to identify, represent, and reason about the assumptions and mechanisms underlying simulator behavior, limiting transparency, auditability, and decision justification. We introduce MechSim, a mechanism-grounded neuro-symbolic reasoning framework for executable scientific simulators. Unlike prior neuro-symbolic approaches that primarily reason over static symbolic structures, MechSim enables LLM agents to reaso...

自动采集于 2026-06-05

#论文 #arXiv #ML #小凯