Asynchronous Neural Networks for Learning in Graphs

TL;DR

This paper introduces asynchronous message passing (AMP) for graph neural networks, enhancing expressiveness and scalability over traditional synchronous models, with theoretical proofs and empirical validation on graph classification tasks.

Contribution

The paper proposes AMP, a novel asynchronous framework for GNNs, demonstrating its ability to simulate synchronous models and distinguish any graph pair, with improved message propagation.

Findings

AMP can simulate synchronous GNNs

AMP can distinguish any pair of graphs

AMP performs well on graph classification benchmarks

Abstract

This paper studies asynchronous message passing (AMP), a new paradigm for applying neural network based learning to graphs. Existing graph neural networks use the synchronous distributed computing model and aggregate their neighbors in each round, which causes problems such as oversmoothing and limits their expressiveness. On the other hand, AMP is based on the asynchronous model, where nodes react to messages of their neighbors individually. We prove that (i) AMP can simulate synchronous GNNs and that (ii) AMP can theoretically distinguish any pair of graphs. We experimentally validate AMP's expressiveness. Further, we show that AMP might be better suited to propagate messages over large distances in graphs and performs well on several graph classification benchmarks.