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..

Smart Surrogate Losses for Contextual Stochastic Linear Optimization with Robust Constraints

Authors: Hyungki Im, Wyame Benslimane, Paul Grigas

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through experiments on fractional knapsack and alloy production problem instances, we demonstrate that SPO-RC+ effectively handles uncertainty in constraints and that combining truncation with importance reweighting can further improve performance.
Researcher Affiliation Academia Hyungki Im Wyame Benslimane Paul Grigas Department of Industrial Engineering and Operations Research University of California, Berkeley Berkeley, CA 94720 EMAIL
Pseudocode Yes Algorithm 1: SPO-RC+ with Data Truncation and Importance Reweighting
Open Source Code Yes Our code is built on the open source software package Py EPO (Tang and Khalil, 2024). Justification: The code is included in the supplementary material.
Open Datasets No We present computational results from synthetic data on multiple applications. ... We generate synthetic data with polynomial kernel models to describe the dependencies of c and a on x; a detailed explanation of the data generation and model description can be found in Appendix C.2.
Dataset Splits Yes Given a testing dataset Zn = {xi, ci, ai}n i=1, ... We partition the data into four distinct sets: two sets are allocated for split conformal prediction, while the other two are used for training and KMM. ... out-of-sample test set (size 3000) ... trained on datasets of 1000 and 5000 training samples.
Hardware Specification Yes The experiments were conducted on a Mac Book Pro equipped with an Intel chip and 16 GB of RAM.
Software Dependencies No Our code is built on the open source software package Py EPO (Tang and Khalil, 2024).
Experiment Setup Yes All models are trained using the Adam optimizer. The learning rate is set to 1 10 3 for the MSE and RF models, and 4 10 3 for the SPO-RC+ model. Training is conducted for 50 epochs with early stopping based on validation loss to prevent overfitting. ... For KMM, we set the constant B = 1000 and ϵ = m 1 m , where m is the size of the truncated dataset.