EvoOptiGraph: Weakness-Driven Coevolution via Graph-Based Structural Generation for Optimization Modeling
cs.AI
Submitted: 2026-06-25
Updated: 2026-11-05
Code: https://github.com/huggingface/trl
License: http://creativecommons.org/licenses/by/4.0/
The gist: Automating optimization modeling from natural language with large language models (LLMs) faces two key challenges.
Terminology
Abstract
Automating optimization modeling from natural language with large language models (LLMs) faces two key challenges. First, training corpora lack structural diversity. Second, data generation pipelines remain static and decoupled from model learning. To address these challenges, we propose EvoOptiGraph, a novel framework where data and model co-evolve, driven by model weaknesses. EvoOptiGraph represents each mixed-integer linear program (MILP) as an attributed bipartite graph and applies validity-preserving evolutionary operators to generate structurally diverse instances. The evolved graphs are converted into solver code and natural language via deterministic compilation and verified back-translation. Training proceeds in two stages: supervised fine-tuning (SFT) on an initial dataset, followed by reinforcement learning with verifiable rewards (RLVR), where graph-derived weakness signals guide the generation of new instances targeting the model's failures. This forms a closed loop that continuously updates the training distribution. Empirical results on six public datasets show that EvoOptiGraph significantly outperforms larger generalist models, agentic methods, and specialized baselines in accuracy, executability, and generalization. These results demonstrate that targeted data-model coevolution is an effective strategy for improving LLMs on optimization modeling tasks.
Sources
- OptiMUS: Scalable Optimization Modeling with (MI)LP Solvers and Large Language Models
- Autoformulation of Mathematical Optimization Models Using LLMs
- Solver-Informed RL: Grounding Large Language Models for Authentic Optimization Modeling
- Process Supervision-Guided Policy Optimization for Code Generation
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- LLMs for Mathematical Modeling: Towards Bridging the Gap between Natural and Mathematical Languages
- LLMOPT: Learning to Define and Solve General Optimization Problems from Scratch
- From Verifiable Dot to Reward Chain: Harnessing Verifiable Reference-based Rewards for Reinforcement Learning of Open-ended Generation
- Tulu 3: Pushing Frontiers in Open Language Model Post-Training
- Towards Foundation Models for Mixed Integer Linear Programming
- Constructing Industrial-Scale Optimization Modeling Benchmark
- DeepSeek-V3 Technical Report
- Evolution of Heuristics: Towards Efficient Automatic Algorithm Design Using Large Language Model
- OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling
- Qwen2.5 Technical Report
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- BPP-Search: Enhancing Tree of Thought Reasoning for Mathematical Modeling Problem Solving
- A Survey of Optimization Modeling Meets LLMs: Progress and Future Directions
- Qwen3 Technical Report
- VLA-R1: Enhancing Reasoning in Vision-Language-Action Models
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents