Graph neural networks and non-commuting operators

Authors: Mauricio Velasco, Kaiying O'Hare, Bernardo Rychtenberg, Soledad Villar

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate our theoretical results with simple experiments on synthetic and realworld data.
Researcher Affiliation Collaboration Mauricio Velasco Departamento de Informática Universidad Católica del Uruguay Montevideo, Uruguay mauricio.velasco@ucu.edu.uy Kaiying O Hare Departament of Applied Mathematics and Statistics Johns Hopkins University Baltimore, Maryland kohare3@jh.edu Bernardo Rychtenberg Departamento de Informática Universidad Católica del Uruguay Montevideo, Uruguay bernardo.rychtenberg@ucu.edu.uy Soledad Villar Departament of Applied Mathematics and Statistics Johns Hopkins University Baltimore, Maryland svillar3@jhu.edu
Pseudocode No The paper describes a training procedure but does not provide it in a structured pseudocode or algorithm block.
Open Source Code Yes 1Code available: https://github.com/Kkylie/Gt NN_weighted_circulant_graphs and https://github.com/mauricio-velasco/operator Networks
Open Datasets Yes We use the publicly available Movie Lens 100k database, a collection of movie ratings given by a set of 1000 users [39] to 1700 movies.
Dataset Splits No We train our model with 800 training data I and test it on 200 testing data Itest. The paper does not explicitly mention a validation split.
Hardware Specification Yes Running these experiments took a few hours on a regular laptop (just CPU).
Software Dependencies No The paper mentions using ADAM for training but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes We use MSE loss, and use ADAM with learning rate 0.01, β1 = 0.9 and β2 = 0.999 to train our models. For all four models, we set the non-commutative polynomial h(T1, T2) to be any polynomial of degree at most d = 3.