Guardado en:
Detalles Bibliográficos
Autores principales: Shi, Cheng, Pan, Liming, Hu, Hong, Dokmanić, Ivan
Formato: Preprint
Publicado: 2022
Materias:
Acceso en línea:https://arxiv.org/abs/2212.13069
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910305145061376
author Shi, Cheng
Pan, Liming
Hu, Hong
Dokmanić, Ivan
author_facet Shi, Cheng
Pan, Liming
Hu, Hong
Dokmanić, Ivan
contents Graph neural networks (GNNs) excel in modeling relational data such as biological, social, and transportation networks, but the underpinnings of their success are not well understood. Traditional complexity measures from statistical learning theory fail to account for observed phenomena like the double descent or the impact of relational semantics on generalization error. Motivated by experimental observations of ``transductive'' double descent in key networks and datasets, we use analytical tools from statistical physics and random matrix theory to precisely characterize generalization in simple graph convolution networks on the contextual stochastic block model. Our results illuminate the nuances of learning on homophilic versus heterophilic data and predict double descent whose existence in GNNs has been questioned by recent work. We show how risk is shaped by the interplay between the graph noise, feature noise, and the number of training labels. Our findings apply beyond stylized models, capturing qualitative trends in real-world GNNs and datasets. As a case in point, we use our analytic insights to improve performance of state-of-the-art graph convolution networks on heterophilic datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2212_13069
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Homophily modulates double descent generalization in graph convolution networks
Shi, Cheng
Pan, Liming
Hu, Hong
Dokmanić, Ivan
Machine Learning
Disordered Systems and Neural Networks
Graph neural networks (GNNs) excel in modeling relational data such as biological, social, and transportation networks, but the underpinnings of their success are not well understood. Traditional complexity measures from statistical learning theory fail to account for observed phenomena like the double descent or the impact of relational semantics on generalization error. Motivated by experimental observations of ``transductive'' double descent in key networks and datasets, we use analytical tools from statistical physics and random matrix theory to precisely characterize generalization in simple graph convolution networks on the contextual stochastic block model. Our results illuminate the nuances of learning on homophilic versus heterophilic data and predict double descent whose existence in GNNs has been questioned by recent work. We show how risk is shaped by the interplay between the graph noise, feature noise, and the number of training labels. Our findings apply beyond stylized models, capturing qualitative trends in real-world GNNs and datasets. As a case in point, we use our analytic insights to improve performance of state-of-the-art graph convolution networks on heterophilic datasets.
title Homophily modulates double descent generalization in graph convolution networks
topic Machine Learning
Disordered Systems and Neural Networks
url https://arxiv.org/abs/2212.13069