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..
Stochastic Optimization in Semi-Discrete Optimal Transport: Convergence Analysis and Minimax Rate
Authors: Ferdinand Genans, Antoine Godichon-Baggioni, François-Xavier Vialard, Olivier Wintenberger
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 6 Numerical experiments. In this section, we numerically verify our convergence rate guarantees through various examples. All experiments demonstrating convergence rates were repeated 20 times, and the error plots represent the averaged errors. |
| Researcher Affiliation | Academia | Ferdinand Genans1 Antoine Godichon-Baggioni1 François-Xavier Vialard2 Olivier Wintenberger1,3 Sorbonne Université, CNRS, LPSM1 Université Gustave Eiffel, CNRS, LIGM2 Wolfgang Pauli Institute3 EMAIL EMAIL |
| Pseudocode | Yes | Algorithm 1 Projected Stochastic Gradient Descent (PSGD) |
| Open Source Code | Yes | All the code to reproduce our experiments is provided and attached in the supplementary materials. |
| Open Datasets | No | In this section, we numerically verify our convergence rate guarantees through various examples. ... For each example, we generate g randomly, and approximate the associated Laguerre cell measures µ(Lc i(g )). We then fix wi = µ(Lc i(g )). |
| Dataset Splits | No | For each example, we generate g randomly, and approximate the associated Laguerre cell measures µ(Lc i(g )). We then fix wi = µ(Lc i(g )), such that g is optimal by the first-order condition. |
| Hardware Specification | No | All our experiments can be run on any modern computer, even without needing a GPU. |
| Software Dependencies | No | The paper does not explicitly mention specific software dependencies or their version numbers in the main text or supplementary materials. No information found in the NeurIPS Paper Checklist either. |
| Experiment Setup | Yes | We set the learning rate to γ1 = Diam(C), as suggested by the analysis in Theorem 3.2. The step decay parameter b was set to 3/4, unless stated otherwise. ... The Laguerre cells are estimated with 109 samples. ... Adam: β1 = 0.9, β2 = 0.999, = 0.001 S-Adam: β1,k = 0.9, β2,k = 1 - 0.9/k, = 10^-3 |