[Paper Review] Artificial Intelligence for Operations Research: Revolutionizing the Operations Research Process

Zhenan Fan, Bissan Ghaddar+4 more|arXiv (Cornell University)|Jan 6, 2024

Scheduling and Optimization Algorithms4 citations

TL;DR

This paper proposes a comprehensive integration of artificial intelligence (AI) into the operations research (OR) pipeline, enhancing parameter generation, model formulation, and optimization through techniques like predict-then-optimize, large language models for natural language-to-math translation, and AI-driven algorithm selection. The key contribution is a unified framework that leverages AI to automate and improve each stage of OR, significantly boosting solution quality and efficiency across complex decision-making problems.

ABSTRACT

The rapid advancement of artificial intelligence (AI) techniques has opened up new opportunities to revolutionize various fields, including operations research (OR). This survey paper explores the integration of AI within the OR process (AI4OR) to enhance its effectiveness and efficiency across multiple stages, such as parameter generation, model formulation, and model optimization. By providing a comprehensive overview of the state-of-the-art and examining the potential of AI to transform OR, this paper aims to inspire further research and innovation in the development of AI-enhanced OR methods and tools. The synergy between AI and OR is poised to drive significant advancements and novel solutions in a multitude of domains, ultimately leading to more effective and efficient decision-making.

Motivation & Objective

• To investigate how AI can enhance each stage of the operations research (OR) process, including parameter generation, model formulation, and optimization.
• To identify synergies between AI and OR that enable automated, accurate, and efficient decision-making in complex domains.
• To address limitations in traditional OR workflows by introducing AI-driven methods that reduce reliance on manual expert input and improve model adaptability.
• To explore feedback loops between OR stages—such as using optimization outcomes to refine model formulations or parameter predictions—through AI.
• To advance the state of the art in algorithm selection and configuration by leveraging AI to predict optimal software and parameter settings for given problem instances.

Proposed method

• Employing the predict-then-optimize (SPO) framework to use AI models for predicting uncertain parameters and optimizing decisions simultaneously, improving both accuracy and feasibility.
• Using large language models (LLMs) such as ChatGPT and Llama to automatically translate natural language problem descriptions into formal mathematical models, reducing the need for manual formulation.
• Applying AI techniques like deep learning and reinforcement learning to handle high-dimensional, dynamic, and complex data structures in parameter generation and model construction.
• Implementing AI-based performance modeling for algorithm runtime prediction, enabling prediction of optimal solver performance on unseen problem instances.
• Integrating AI into algorithm selection and automatic configuration by modeling the relationship between problem features and solver performance across tools like Gurobi, CPLEX, and OptVerse.
• Designing similarity metrics between problem instances to improve generalization and robustness of AI models in OR applications, particularly in runtime prediction and configuration.

Experimental results

Research questions

• RQ1 How can AI be systematically integrated into each stage of the operations research pipeline to improve model accuracy and computational efficiency?
• RQ2 To what extent can large language models accurately translate natural language problem descriptions into formal mathematical optimization models?
• RQ3 Can AI-driven parameter prediction frameworks, such as predict-then-optimize, improve solution quality while ensuring constraint feasibility in real-world applications?
• RQ4 How can feedback from optimization outcomes inform and refine earlier stages, such as model formulation or parameter generation, through AI?
• RQ5 What are the challenges and opportunities in using AI for unified software selection and automatic algorithm configuration in OR, and how can performance prediction models be made robust and generalizable?

Key findings

• AI models can outperform traditional heuristics and even state-of-the-art commercial solvers like Gurobi and CPLEX on mixed-integer programming (MIP) problems with hundreds or thousands of variables, particularly when leveraging learned policies from expensive environments.
• The predict-then-optimize framework enables AI to simultaneously predict uncertain parameters and optimize decisions, improving solution quality and feasibility, especially when constraints are relaxed to avoid infeasibility.
• Large language models can successfully convert natural language descriptions into mathematically valid optimization formulations, reducing the barrier to entry for domain experts without deep OR expertise.
• AI-based performance models for algorithm runtime prediction show strong empirical results, though their generalization is limited by the lack of a well-defined similarity metric between problem instances.
• Feedback loops between OR stages—such as using optimization results to refine model formulations or parameter predictions—can significantly enhance solution quality and system adaptability.
• Unified software selection and automatic configuration using AI remain challenging due to the black-box nature of solvers and complex parameter spaces, but AI-driven prediction models offer a promising path forward.