HyperGraphX: Graph Transductive Learning with Hyperdimensional Computing and Message Passing

ABSTRACT We present a novel algorithm, HyperGraphX, that marries graph convolution with binding and bundling operations in hyperdimensional computing for transductive graph learning. For prediction accuracy HyperGraphX outperforms major and popular graph neural network implementations as well as state-of-the-art hyperdimensional computing implementations for a collection of homophilic graphs and heterophilic graphs. Compared with the most accurate learning methodologies we have tested, on the same target GPU platform, HyperGraphX is on average 9561.0 and 144.5 times faster than GCNII, a graph neural network implementation and HDGL, a hyperdimensional computing implementation, respectively. As the majority of the learning operates on binary vectors, we expect outstanding energy performance of Hyper-GraphX on neuromorphic and emerging process-in-memory devices. Index T erms-- node classification, hyperdimensional computing, energy efficiency, graph convolution 1. INTRODUCTION In transductive learning both labeled and unlabeled instances are present during training.