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..
Nonlinearly Preconditioned Gradient Methods: Momentum and Stochastic Analysis
Authors: Konstantinos Oikonomidis, Jan Quan, Panagiotis Patrinos
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In numerical simulations we show that the proposed methods perform well on a variety of machine learning problems, including neural network training. ... We first consider some neural network experiments. ... Image classification on the MNIST dataset [9]. ... Image classification on the Cifar10 dataset [21]. ... Matrix factorization ... We provide more details about the implementation along with further experimental results in Appendix F. |
| Researcher Affiliation | Academia | Konstantinos Oikonomidis ESAT-STADIUS & Leuven.AI KU Leuven EMAIL Jan Quan ESAT-STADIUS & Leuven.AI KU Leuven EMAIL Panagiotis Patrinos ESAT-STADIUS & Leuven.AI KU Leuven EMAIL |
| Pseudocode | Yes | Algorithm 1 Nonlinearly preconditioned gradient method with momentum (m-NPGM) |
| Open Source Code | Yes | The code for reproducing the experiments is publicly available2. 2https://github.com/JanQ/nonlin-prec-mom-stoch |
| Open Datasets | Yes | Image classification on the MNIST dataset [9]. ... Image classification on the Cifar10 dataset [21]. ... We use the Movielens 100K dataset [13] |
| Dataset Splits | No | The paper mentions datasets like MNIST, Cifar10, and Movielens 100K, which have standard splits, but it does not explicitly state the percentages, sample counts, or specific methodology for training/test/validation splits used in their experiments. It only mentions batch sizes (e.g., 'batch size of 256', 'batch size 128'). |
| Hardware Specification | Yes | All neural networks experiments were carried out on an NVIDIA P100 GPU in an internal cluster. ... The matrix factorization experiments were run on a Intel Core i7-11700 @ 2.50GHz CPU |
| Software Dependencies | No | The paper mentions using 'standard Py Torch [38] implementations' but does not specify a version number for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | For all the methods we used fixed stepsizes. ... We tuned the stepsizes by performing an adaptive gridsearch ... batch size of 256. ... We use batch size 128 for all methods. ... We choose β = 0.9 for all methods ... We set the learning rate by performing a parameter sweep over {5, 1, 0.5, 0.1, 0.05, 0.01, 0.005}. ... For our method we use β = 0.9 and stepsize γ = 2, but consider the method generated by ϕ(x) = λ(cosh( x ) 1) with λ = 100 ... For GDm we use momentum parameter β = 0.9 and γ = 1/300. |