Quantifying the Alignment of Graph and Features in Deep Learning

Abstract

We show that the classification performance of graph convolutional networks (GCNs) is related to the alignment between features, graph, and ground truth, which we quantify using a subspace alignment measure (SAM) corresponding to the Frobenius norm of the matrix of pairwise chordal distances between three subspaces associated with features, graph, and ground truth. The proposed measure is based on the principal angles between subspaces and has both spectral and geometrical interpretations. We showcase the relationship between the SAM and the classification performance through the study of limiting cases of GCNs and systematic randomizations of both features and graph structure applied to a constructive example and several examples of citation networks of different origins. The analysis also reveals the relative importance of the graph and features for classification purposes.

arXiv commentPublished in IEEE Transactions on Neural Networks and Learning Systems; Date of Publication: 11 January 2021

References in corpus (10)

• The structure and function of complex networks
• Deep Learning in Neural Networks: An Overview
• A Comprehensive Survey on Graph Neural Networks
• Inductive Representation Learning on Large Graphs
• Geometric deep learning: going beyond Euclidean data
• Deep Convolutional Networks on Graph-Structured Data
• Simplifying Graph Convolutional Networks
• Random Walks, Markov Processes and the Multiscale Modular Organization of Complex Networks
• Laplacian Dynamics and Multiscale Modular Structure in Networks
• Learning Multiagent Communication with Backpropagation

Cited by in corpus (3)

• Local dominance unveils clusters in networks
• Structure-Aware DropEdge Towards Deep Graph Convolutional Networks
• Understanding Deep Learning via Decision Boundary