usail-hkust/LLM-MM-Agent

What it solves

MM-Agent addresses the complexity of real-world mathematical modeling, which typically requires a multi-step human process of analyzing unstructured problems, formulating mathematical models, implementing computational solutions, and writing academic reports. It automates this end-to-end workflow to act as a "modeling copilot."

How it works

The system simulates a human expert's workflow through four structured stages:

1. Problem Analysis: Decomposes complex, unstructured problem descriptions into manageable subtasks and identifies constraints.
2. Mathematical Modeling: Translates the problem into mathematical formulations using assumptions and retrieves appropriate strategies from the Hierarchical Mathematical Modeling Library (HMML), a tri-level knowledge hierarchy of domains, subdomains, and 98 method nodes.
3. Computational Solving: Uses an "MLE-Solver" to autonomously generate, execute, and iteratively improve code for algorithms and simulations.
4. Solution Reporting: Summarizes the process and interprets results to generate structured, academic-quality reports.

Who it’s for

It is designed for students and researchers participating in mathematical modeling competitions (such as MCM/ICM) and professionals needing to automate the translation of real-world problems into mathematical solutions.

Highlights

• End-to-End Automation: Covers everything from initial problem analysis to the final paper writing.
• HMML Knowledge Base: Utilizes a specialized hierarchical library for intelligent model selection.
• Iterative Code Generation: Automatically generates and refines code to ensure accurate computational results.
• Academic Reporting: Capable of producing professional-grade reports and visualizations.