Principles for initialization and architecture selection in graph neural networks with ReLU activations

DOI10.1137/23M1600621MaRDI QIDQ6664474FDOQ6664474

Authors: Gage DeZoort, Boris Hanin 

Publication date: 16 January 2025

Published in: SIAM Journal on Mathematics of Data Science (Search for Journal in Brave)

Recommendations

• Theory of graph neural networks: representation and learning
• The logic of graph neural networks
• Activation function design for deep networks: linearity and effective initialisation
• \(k\)-hop graph neural networks

zbMATH Keywords

initializationoversmoothinggraph neural networks

Mathematics Subject Classification ID

Artificial neural networks and deep learning (68T07) Artificial intelligence (68T99)

Cites Work

• Sharp nonasymptotic bounds on the norm of random matrices with independent entries
• Limit of the smallest eigenvalue of a large dimensional sample covariance matrix
• Products of many large random matrices and gradients in deep neural networks
• The Principles of Deep Learning Theory

This page was built for publication: Principles for initialization and architecture selection in graph neural networks with ReLU activations

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6664474)