G-Sim: Generative Simulations with Large Language Models and Gradient-Free Calibration

2025-06-10

AI Summary

Simulation models in critical domains such as healthcare and logistics suffer from poor generalizability, weak causal interpretability, and difficulty in empirical calibration. Method: This paper introduces the first LLM-driven framework for structured modeling and gradient-free empirical calibration, integrating large language models (for iterative causal structure generation), likelihood-free optimization (e.g., CMA-ES), simulation-based inference (SBI), and domain-knowledge injection—enabling parameter estimation and posterior inference for non-differentiable, stochastic simulators. Contribution/Results: Unlike conventional data-driven simulators or uncalibrated LLM-based generation, our framework significantly improves robustness under out-of-distribution conditions and reliability of policy intervention analysis. It achieves high-fidelity, causally interpretable “what-if” reasoning even under small-sample regimes.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Sampling/Simulation-based Search
• Reasoning under Uncertainty: Causality

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Constructing robust simulators is essential for asking"what if?"questions and guiding policy in critical domains like healthcare and logistics. However, existing methods often struggle, either failing to generalize beyond historical data or, when using Large Language Models (LLMs), suffering from inaccuracies and poor empirical alignment. We introduce G-Sim, a hybrid framework that automates simulator construction by synergizing LLM-driven structural design with rigorous empirical calibration. G-Sim employs an LLM in an iterative loop to propose and refine a simulator's core components and causal relationships, guided by domain knowledge. This structure is then grounded in reality by estimating its parameters using flexible calibration techniques. Specifically, G-Sim can leverage methods that are both likelihood-free and gradient-free with respect to the simulator, such as gradient-free optimization for direct parameter estimation or simulation-based inference for obtaining a posterior distribution over parameters. This allows it to handle non-differentiable and stochastic simulators. By integrating domain priors with empirical evidence, G-Sim produces reliable, causally-informed simulators, mitigating data-inefficiency and enabling robust system-level interventions for complex decision-making.

Problem

• Constructing robust simulators for policy guidance
• Addressing inaccuracies in LLM-based simulation methods
• Enabling non-differentiable and stochastic simulator calibration

Innovation

• LLM-driven structural design for simulators
• Gradient-free optimization for parameter calibration
• Simulation-based inference for posterior distributions