Minimal Variance Sampling with Provable Guarantees for Fast Training of Graph Neural Networks

Weilin Cong, Rana Forsati, Mahmut Kandemir, Mehrdad Mahdavi

Research output: Chapter in Book/Report/Conference proceedingConference contribution

6 Scopus citations

Abstract

Sampling methods (e.g., node-wise, layer-wise, or subgraph) has become an indispensable strategy to speed up training large-scale Graph Neural Networks (GNNs). However, existing sampling methods are mostly based on the graph structural information and ignore the dynamicity of optimization, which leads to high variance in estimating the stochastic gradients. The high variance issue can be very pronounced in extremely large graphs, where it results in slow convergence and poor generalization. In this paper, we theoretically analyze the variance of sampling methods and show that, due to the composite structure of empirical risk, the variance of any sampling method can be decomposed intoembedding approximation variance in the forward stage andstochastic gradient variance in the backward stage that necessities mitigating both types of variance to obtain faster convergence rate. We propose a decoupled variance reduction strategy that employs (approximate) gradient information to adaptively sample nodes with minimal variance, and explicitly reduces the variance introduced by embedding approximation. We show theoretically and empirically that the proposed method, even with smaller mini-batch sizes, enjoys a faster convergence rate and entails a better generalization compared to the existing methods.

Original languageEnglish (US)
Title of host publicationKDD 2020 - Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
PublisherAssociation for Computing Machinery
Pages1393-1403
Number of pages11
ISBN (Electronic)9781450379984
DOIs
StatePublished - Aug 23 2020
Event26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2020 - Virtual, Online, United States
Duration: Aug 23 2020Aug 27 2020

Publication series

NameProceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Conference

Conference26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2020
Country/TerritoryUnited States
CityVirtual, Online
Period8/23/208/27/20

All Science Journal Classification (ASJC) codes

  • Software
  • Information Systems

Fingerprint

Dive into the research topics of 'Minimal Variance Sampling with Provable Guarantees for Fast Training of Graph Neural Networks'. Together they form a unique fingerprint.

Cite this