Idea
An AI-driven optimization platform using large language models to solve complex nonlinear combinatorial problems for enterprises and researchers.
Research Paper
Core Innovation
This paper presents AutoCO, the first end-to-end automated constraint optimization framework that uses large language models to learn and evolve constraint relaxation strategies. It uniquely combines bidirectional coevolution of evolutionary algorithms and Monte Carlo Tree Search to effectively explore and refine solutions in fragmented, multi-modal spaces, surpassing prior passive LLM-based validators.
Market Size (TAM)
$2–10B TAM for combinatorial optimization software; $1–2B SAM from logistics, manufacturing, and AI research sectors. Driven by increasing demand for automated optimization and AI integration.
Potential Customers & Pain Points
- Enterprises facing complex nonlinear combinatorial optimization challenges
- Researchers needing scalable automated optimization tools
- Software developers lacking adaptive constraint relaxation methods
Business Model
Subscription-based SaaS platform offering API access and enterprise licenses with customization and support services.
Competitive Landscape
- Gurobi
- CPLEX
- Google OR-Tools
Implementation Challenges
- Integration complexity with existing systems
- Computational resource demands
- Adoption resistance due to trust in traditional methods
Validation Strategy
- Benchmark AutoCO against leading optimization solvers on standard NCOP datasets
- Pilot deployments with logistics and manufacturing firms
- Collect user feedback to refine LLM-driven strategy generation
Research Paper Overview
Learn to Relax with Large Language Models: Solving Nonlinear Combinatorial Optimization Problems via Bidirectional Coevolution
Summary
This paper introduces AutoCO, an automated constraint optimization method that leverages large language models to generate dynamic constraint relaxation strategies for nonlinear combinatorial optimization problems. It integrates evolutionary algorithms for local refinement with Monte Carlo Tree Search for global strategy exploration, achieving a balance between intensification and diversification in complex solution spaces. Experiments on three benchmarks demonstrate AutoCO's superior performance over existing methods.