Research map / Large language models / Agents, code and math
VeriSimpl: Robust Optimization Modeling from Natural Language using Simplification-based Verification
Sumaya Abdul Rahman, Seckhen Ariel Andrade Cuellar, Ghani Raissov, Mohammad Raza
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add simplification-based verification and candidate selection to the base LLM | GPT-4o · average over 4 optimization benchmarks | accuracy: 56.6 -> 65.5 | Improved |
| add simplification-based verification to the base LLM | DeepSeek-R1 · average over 4 optimization benchmarks | accuracy: 62.5 -> 72.8 | Improved |
| use VeriSimpl instead of OptiMUS multi-agent pipeline | GPT-4o · IndOR | accuracy: 17.7 -> 45.8 | Improved |
| add full-variable verification, single-variable, and type verification to constraint verification (full VeriSimpl vs A-Cons) (ablation) | VeriSimpl · average accuracy, GPT-4o | accuracy: 63.9 -> 65.5 | Improved |
| add all verification signals vs only full-variable verification (ablation) | VeriSimpl · average accuracy, GPT-4o | accuracy: 62.2 -> 65.5 | Improved |
| add the other verification signals to singleton-variable verification (ablation) | VeriSimpl · average accuracy, DeepSeek-R1 | accuracy: 68.9 -> 72.5 | Improved |
| add type-based verification and refinement (ablation) | VeriSimpl · average accuracy, DeepSeek-R1 | accuracy: 72.3 -> 72.5 | No clear effect |
| add type-based verification and refinement (ablation) | VeriSimpl · average accuracy, GPT-4o | accuracy: 64.2 -> 65.5 | Improved |
| require all verification queries to pass (full self-verification) as a confidence signal | VeriSimpl · average over datasets, GPT-4o | precision: 65.5 accuracy -> 91.5 precision (coverage 34.2) | Improved |
| use R1 instead of GPT-4o as the base for self-verification | VeriSimpl · average over datasets | precision: 91.5 -> 78.5 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OptiMUS — OptiMUS: Scalable Optimization Modeling with (MI)LP Solvers and Large Language Models (ICML 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling (ICML 2025 · similarity 0.91)
- Autoformulation of Mathematical Optimization Models Using LLMs (ICML 2025 · similarity 0.90)
- LLMOPT: Learning to Define and Solve General Optimization Problems from Scratch (ICLR 2025 · similarity 0.90)
- Automated Reformulation of Robust Optimization via Memory-Augmented Large Language Models (NeurIPS 2026 · similarity 0.89)
- PEARL: Solver-in-the-Loop Interactive Optimization Modeling from Natural Language (NeurIPS 2026 · similarity 0.88)
- Solver-Informed RL: Grounding Large Language Models for Authentic Optimization Modeling (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
