Timezone: »
We propose and study a practical graph embedding that in expectation is able to distinguish all non-isomorphic graphs and can be computed in polynomial time. The embedding is based on Lovász' characterization of graph isomorphism through an infinite dimensional vector of homomorphism counts. Recent work has studied the expressiveness of graph embeddings by comparing their ability to distinguish graphs to that of the Weisfeiler-Leman hierarchy. While previous methods have either limited expressiveness or are computationally impractical, we devise efficient sampling-based alternatives that are maximally expressive in expectation. We empirically evaluate our proposed embeddings and show competitive results on several benchmark graph learning tasks.
Author Information
Maximilian Thiessen (TU Wien)
Pascal Welke (University of Bonn)
Thomas Gärtner (TU Wien)
More from the Same Authors
-
2022 : Generalized Laplacian Positional Encoding for Graph Representation Learning »
Sohir Maskey · Ali Parviz · Maximilian Thiessen · Hannes Stärk · Ylli Sadikaj · Haggai Maron -
2022 Poster: Active Learning of Classifiers with Label and Seed Queries »
Marco Bressan · Nicolò Cesa-Bianchi · Silvio Lattanzi · Andrea Paudice · Maximilian Thiessen -
2021 Poster: Active Learning of Convex Halfspaces on Graphs »
Maximilian Thiessen · Thomas Gaertner