Evolving-Graph Gaussian Processes
This addresses a limitation in graph-based machine learning for domains with evolving structures, though it appears incremental as an extension of existing static graph Gaussian Processes.
The authors tackled the problem of applying Gaussian Processes to dynamic graph-structured data by proposing evolving-Graph Gaussian Processes (e-GGPs), which learn transition functions and neighborhood kernels to model connectivity changes over time, and demonstrated its benefits over static methods in time-series regression tasks.
Graph Gaussian Processes (GGPs) provide a data-efficient solution on graph structured domains. Existing approaches have focused on static structures, whereas many real graph data represent a dynamic structure, limiting the applications of GGPs. To overcome this we propose evolving-Graph Gaussian Processes (e-GGPs). The proposed method is capable of learning the transition function of graph vertices over time with a neighbourhood kernel to model the connectivity and interaction changes between vertices. We assess the performance of our method on time-series regression problems where graphs evolve over time. We demonstrate the benefits of e-GGPs over static graph Gaussian Process approaches.