Density of states for fast embedding node-attributed graphs

Abstract Given a node-attributed graph, how can we efficiently represent it with few numerical features that expressively reflect its topology and attribute information? We propose A-DOGE , for attributed DOS-based graph embedding, based on density of states (DOS, a.k.a. spectral density) to tackle this problem. is designed fulfill long desiderata desirable characteristics. Most notably, capitalizes efficient approximation algorithms DOS, extend blend in node labels attributes the first time, making fast scalable large graphs databases. Being entire eigenspectrum capture structural properties at multiple (“glocal”) scales. Moreover, unsupervised (i.e., agnostic any specific objective) lends itself various interpretations, which makes suitable exploratory mining tasks. Finally, processes each independent others, amenable streaming settings as well parallelization. Through extensive experiments, show efficacy efficiency analysis classification tasks, where significantly outperforms baselines achieves competitive performance modern supervised GNNs, while achieving best trade-off between accuracy runtime.