Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Enforcing Hard Linear Constraints in Deep Learning Models with Decision Rules
Authors: Gonzalo E. Constante, Hao Chen, Can Li
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results on benchmark regression tasks show that our method consistently satisfies constraints while maintaining competitive accuracy and low inference latency. We evaluate our framework on two end-to-end constrained optimization learning tasks: |
| Researcher Affiliation | Academia | Gonzalo E. Constante Davidson School of Chemical Engineering Purdue University EMAIL Hao Chen Davidson School of Chemical Engineering Purdue University EMAIL Can Li Davidson School of Chemical Engineering Purdue University EMAIL |
| Pseudocode | No | The paper describes the proposed framework and its components using mathematical formulations and descriptive text, but it does not include explicitly structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The implementation of all the methods can be found at this Git Hub repository https://github. com/li-group/Decision Rule Net. |
| Open Datasets | Yes | DC Optimal Power Flow (DC-OPF): The task is to predict the optimal generation output for varying demands. ... [30] Sogol Babaeinejadsarookolaee, Adam Birchfield, Richard D. Christie, Carleton Coffrin, Christopher De Marco, Ruisheng Diao, Michael Ferris, Stephane Fliscounakis, Scott Greene, Renke Huang, Cedric Josz, Roman Korab, Bernard Lesieutre, Jean Maeght, Terrence W. K. Mak, Daniel K. Molzahn, Thomas J. Overbye, Patrick Panciatici, Byungkwon Park, Jonathan Snodgrass, Ahmad Tbaileh, Pascal Van Hentenryck, and Ray Zimmerman. The Power Grid Library for benchmarking AC optimal power flow algorithms, 2021. Portfolio Optimization: The task is to predict optimal portfolio allocations under uncertain maturity and rating of bonds[31]... [31] Mark Rubinstein. Markowitz s" portfolio selection": A fifty-year retrospective. The Journal of finance, 57(3):1041 1045, 2002. |
| Dataset Splits | No | For each problem, we generated 100 samples of the input x uniformly sampled from the uncertainty set X. The reported metrics are averaged over these test samples. For all methods, hyperparameters were selected based on a validation set and fixed across problem sizes to ensure fairness. This describes test data generation and mentions a validation set, but does not provide specific training/test/validation split percentages or sample counts. |
| Hardware Specification | Yes | The ground truths for all instances were solved using the Gurobi solver on an Apple M2 Pro CPU and 32GB of RAM. All neural networks were trained on a NVIDIA T4 Tensor Core GPU with parallelization using Py Torch. |
| Software Dependencies | No | The ground truths for all instances were solved using the Gurobi solver on an Apple M2 Pro CPU and 32GB of RAM. All neural networks were trained on a NVIDIA T4 Tensor Core GPU with parallelization using Py Torch. The specific version numbers for PyTorch, Gurobi, and OSQP are not provided. |
| Experiment Setup | Yes | The following configurations and hyperparameters are fixed throughout all experiments and all methods, based on preliminary experimentation to confirm the proper convergence of training. Epochs: 300 for DCOPF, 500 for portfolio optimization Optimizer: Adam Learning rate: 10 4 Batch size: 64 Hidden layer number: 2 Hidden layer size: 256 Activation: Re LU Batch normalization: True Feasibility method stopping tolerance: 10 4 Feasibility method maximum iterations: 300 DC3 correction procedure momentum: 0.5 DC3 correction learning rate: 10 4 |