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..

Generalization Bounds for Rank-sparse Neural Networks

Authors: Antoine Ledent, Rodrigo Alves, Yunwen Lei

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present some experiments to demonstrate that our bound can successfully capture the low rank structure in the weights to improve upon competing bounds.
Researcher Affiliation Academia Antoine Ledent School of Computing and Information Systems (SCIS) Singapore Management University (SMU) EMAIL; Rodrigo Alves Department of Applied Mathematics Czech Technical University in Prague (CTU) EMAIL; Yunwen Lei Department of Mathematics The University of Hong Kong EMAIL
Pseudocode No The paper describes methods in mathematical notation and prose, but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [No]
Open Datasets Yes We present results for both DNNs on the MNIST dataset and CNNs on the CIFAR-10 dataset.
Dataset Splits No The paper mentions using MNIST and CIFAR-10 datasets but does not explicitly detail the training, validation, or test splits. It refers to a margin and accuracy for evaluation.
Hardware Specification Yes All experiments were run with 128 CPUs with 500GB ram and DGX A100.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes We set the number of non-linear hidden layers to L = 3 and varied the width of the hidden layers wβ„“in t300, 400, 500, 700u to explore different regimes. ... We selected the regularization parameter from t10 4, 5 Γ— 10 4, 10 3, ..., 102u and analyzed models with the highest weight decay that achieved an accuracy of at least 0.9. We then maximize the margin Ξ³ subject to IΞ³,X ≀ 0.1.