Self-Supervised Primal-Dual Learning for Constrained Optimization

Authors: Seonho Park, Pascal Van Hentenryck

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that, on a set of nonlinear optimization benchmarks, PDL typically exhibits negligible constraint violations and minor optimality gaps, and is remarkably close to the ALM optimization.
Researcher Affiliation Academia H. Milton Stewart School of Industrial & Systems Engineering, Georgia Institute of Technology seonho.park@gatech.edu, pascal.vanhentenryck@isye.gatech.edu
Pseudocode Yes Algorithm 1 Primal-Dual Learning (PDL)
Open Source Code No The paper does not provide a direct link or explicit statement for the open-source code of the described methodology (PDL). It only provides a link for a baseline method: '3https://github.com/locuslab/DC3'
Open Datasets Yes Table 3 reports the performance results of the case56 and case118 in PGLib (Babaeinejadsarookolaee et al. 2019).
Dataset Splits Yes Overall, 10,000 instances were generated and split into training/testing/validation datasets with the ratio of 10:1:1.
Hardware Specification Yes The implementation is based on Py Torch and the training was conducted using a Tesla RTX6000 GPU on a machine with Intel Xeon 2.7GHz.
Software Dependencies Yes Gurobi 9.5.2 was used as the optimization solver to produce the instance data for the supervised baselines, and also served as the reference for computing optimality gaps.
Experiment Setup Yes For training the models, the Adam optimizer (Kingma and Ba 2014) with the learning rate of 1e-4 was used. Other hyperparameters of PDL and the baselines were tuned using a grid search.