Learning on Random Balls is Sufficient for Estimating (Some) Graph Parameters

Abstract

Theoretical analyses for graph learning methods often assume a complete observation of the input graph. Such an assumption might not be useful for handling any-size graphs due to the scalability issues in practice. In this work, we develop a theoretical framework for graph classification problems in the partial observation setting (i.e., subgraph samplings). Equipped with insights from graph limit theory, we propose a new graph classification model that works on a randomly sampled subgraph and a novel topology to characterize the representability of the model. Our theoretical framework contributes a theoretical validation of mini-batch learning on graphs and leads to new learning-theoretic results on generalization bounds as well as size-generalizability without assumptions on the input.

The manuscript is accepted as a poster presentation at NeurIPS 2021. This ArXiv version includes the Appendix

References in corpus (5)

• Interaction Networks for Learning about Objects, Relations and Physics
• TUDataset: A collection of benchmark datasets for learning with graphs
• Revisiting Graph Neural Networks: All We Have is Low-Pass Filters
• Graph Homomorphism Convolution
• Generalization bounds for graph convolutional neural networks via Rademacher complexity