EquivaMap: Leveraging LLMs for Automatic Equivalence Checking of Optimization Formulations
Authors: Haotian Zhai, Connor Lawless, Ellen Vitercik, Liu Leqi
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To evaluate our approach, we construct Equiva Formulation, the first open-source dataset of equivalent optimization formulations, generated by applying transformations such as adding slack variables or valid inequalities to existing formulations. Empirically, Equiva Map significantly outperforms existing methods, achieving substantial improvements in correctly identifying formulation equivalence.1 |
| Researcher Affiliation | Academia | 1The University of Texas at Austin, TX, USA 2Stanford University, CA, USA. |
| Pseudocode | Yes | Algorithm 1 Equiva Map |
| Open Source Code | Yes | 1The code and datasets are available at https: //github.com/Humain Lab/Equiva Map and https://huggingface.co/datasets/humainlab/ Equiva Formulation. |
| Open Datasets | Yes | To evaluate our approach, we construct Equiva Formulation, the first open-source dataset of equivalent optimization formulations... 1The code and datasets are available at https: //github.com/Humain Lab/Equiva Map and https://huggingface.co/datasets/humainlab/ Equiva Formulation. We construct Equiva Formulation based on the NLP4LP dataset (Ahmadi Teshnizi et al., 2024). |
| Dataset Splits | No | The paper does not explicitly provide training/test/validation dataset splits for reproducing the experiment. It describes generating a dataset and then evaluating methods on it without specifying how that dataset itself is split for learning or evaluation in a traditional train/test sense. It measures accuracy on the overall generated dataset. |
| Hardware Specification | No | The paper does not explicitly mention specific hardware (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper mentions using 'GPT-4' and 'GPT-4o' as LLMs and 'Gurobi' as an MILP solver, but it does not specify explicit version numbers for these software components or any other libraries. |
| Experiment Setup | Yes | We use GPT-4 (Achiam et al., 2023) as the mapping finder in Equiva Map... We set K = 3, and report the accuracy as the percentage of paired formulations α and α that are correctly identified as equivalent or nonequivalent |