LLM-Guided Evolution of Symbolic Acceleration Inequality Families for MILP
The paper introduces a framework that uses Large Language Models (LLMs) and evolutionary algorithms to automatically discover reusable acceleration cuts for Mixed-Integer Linear Programming (MILP) models. By reasoning over the model and its description, the system generates cuts that speed up solving times across different problem instances. This means AI can learn structural knowledge from natural language descriptions to create practical, high-impact solver enhancements.