Compact Optimality Verification for Optimization Proxies
Authors: Wenbo Chen, Haoruo Zhao, Mathieu Tanneau, Pascal Van Hentenryck
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results are provided to highlight the benefits of the compact optimality verification formulation. Extensive experiments show the compact formulation, together with the primal heuristic, can effectively verify optimization proxies for large-scale DC-OPF problems and knapsack problems. |
| Researcher Affiliation | Academia | 1H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, USA 2NSF Artificial Intelligence Research Institute for Advances in Optimization (AI4OPT), USA. Correspondence to: Wenbo Chen <wenbo.chen@gatech.edu>. |
| Pseudocode | Yes | Algorithm 1 Projected Gradient Attack with Value Function Approximation (PGA-VFA) and Algorithm 2 Hypersimplex Projection via Binary Search |
| Open Source Code | No | The paper discusses the tools and libraries used (e.g., PyTorch, Gurobi) but does not provide an explicit statement about releasing its own source code or a link to a repository for the methodology described in the paper. |
| Open Datasets | Yes | The optimality verification for DCOPF proxies is evaluated over IEEE 57-/118-/300-bus and Pegase 1354-bus test cases from the PGLib library (Babaeinejadsarookolaee et al., 2019). |
| Dataset Splits | No | The paper describes how instances are generated for evaluation and that optimization proxies are trained, but it does not provide explicit training, validation, and test dataset splits in terms of percentages, absolute counts, or references to predefined splits for the experimental setup. |
| Hardware Specification | Yes | Experiments are conducted on dual Intel Xeon 6226@2.7GHz machines running Linux on the cluster. |
| Software Dependencies | Yes | All verification problems are solved with Gurobi 10.0 (Gurobi Optimization, LLC, 2023) using 16 threads... |
| Experiment Setup | Yes | PGA-VFA is executed in parallel, across 200 threads, using the acceleration techniques of Section 5. The initial step size is 10^-3, and is reduced by a factor 10 if no improvement is recorded over 10 iterations. Finally, PGA-VFA is stopped if no improved solution is found after 20 consecutive iterations, or a maximum of 500 iterations is reached. All verification problems are solved with Gurobi 10.0 (Gurobi Optimization, LLC, 2023) using 16 threads and a 6-hour time limit. |