Distributed change point detection in streaming manifold-valued signals over graphs

Description

Signal processing methods over graphs and networks have recently been proposed to detect change points occurring in localized communities of nodes. Nevertheless, all these methods are mostly limited to time series data in Euclidean spaces. In this paper, we devise a distributed change point detection method for streaming manifold-valued signals over graphs. This framework combines a local test statistic at each node to account for the geometry of the data residing on a Riemannian manifold, with a fully distributed graph filter that incorporates information on network topology. This significantly improves the detection of change points in unknown communities of networks.

Abstract

International audience

Additional details