BriefGPT.xyz
Dec, 2020
一种基于PAC-Bayesian方法的图神经网络泛化界的研究
A PAC-Bayesian Approach to Generalization Bounds for Graph Neural Networks
HTML
PDF
Renjie Liao, Raquel Urtasun, Richard Zemel
TL;DR
本文基于PAC-Bayesian方法推导出了两种主要的图神经网络（GCNs和MPGNNs）的泛化界，进一步显示节点最大度数和权重的谱范数支配了这两种模型的泛化界。
Abstract
In this paper, we derive
generalization bounds
for the two primary classes of
graph neural networks
(GNNs), namely
graph convolutional networks
→