Oral
in
Workshop: Bayesian Decision-making and Uncertainty: from probabilistic and spatiotemporal modeling to sequential experiment design
Graph Classification Gaussian Processes via Hodgelet Spectral Features
Mathieu Alain · So Takao · Bastian Rieck · Xiaowen Dong · Emmanuel Noutahi
Keywords: [ Gaussian Processes ] [ Classification ] [ graphs ] [ Spectral Features ]
The problem of classifying graphs is ubiquitous in machine learning. While it is standard to apply graph neural networks for such tasks,Gaussian processes can also be used for graph classification, by transforming the graph signals into the spectral domain, and using the spectral features as inputs. However, this approach only takes into account signals on the vertices of the graph, whereas some data also support signals on the edges. In this work, we present a Gaussian process-based classification algorithm that can utilise both vertex and/or edges features to help classify graphs. Furthermore, we take advantage of the Hodge decomposition of vertex and edge signals to increase the flexibility to the model, which can be beneficial on some tasks.