DAOpt: Modeling and Evaluation of Data-Driven Optimization under Uncertainty with LLMs

TL;DR

This paper introduces DAOpt, a framework leveraging large language models for uncertain optimization problems, featuring a new dataset, decision-making module, and evaluation environment to improve robustness and out-of-sample feasibility.

Contribution

It presents a novel framework and dataset for applying LLMs to uncertain optimization, integrating multi-agent decision-making and domain knowledge for enhanced modeling.

Findings

Demonstrates improved out-of-sample feasibility and robustness of LLMs in uncertain optimization.

Introduces a new dataset and simulation environment for evaluating LLMs in decision-making under uncertainty.

Enhances LLM capabilities with few-shot learning and domain knowledge from stochastic and robust optimization.

Abstract

Recent advances in large language models (LLMs) have accelerated research on automated optimization modeling. While real-world decision-making is inherently uncertain, most existing work has focused on deterministic optimization with known parameters, leaving the application of LLMs in uncertain settings largely unexplored. To that end, we propose the DAOpt framework including a new dataset OptU, a multi-agent decision-making module, and a simulation environment for evaluating LLMs with a focus on out-of-sample feasibility and robustness. Additionally, we enhance LLMs' modeling capabilities by incorporating few-shot learning with domain knowledge from stochastic and robust optimization.