[2006.10628] Offline detection of change-points within the imply for stationary graph indicators

Obtain a PDF of the paper titled Offline detection of change-points within the imply for stationary graph indicators, by Alejandro de la Concha and a couple of different authors

Obtain PDF
HTML (experimental)

Summary:This paper addresses the issue of segmenting a stream of graph indicators: we purpose to detect adjustments within the imply of a multivariate sign outlined over the nodes of a recognized graph. We suggest an offline methodology that depends on the idea of graph sign stationarity and permits the handy translation of the issue from the unique vertex area to the spectral area (Graph Fourier Rework), the place it’s a lot simpler to unravel. Though the obtained spectral illustration is sparse in actual functions, to one of the best of our information this property has not been sufficiently exploited within the present associated literature. Our change-point detection methodology adopts a mannequin choice method that takes into consideration the sparsity of the spectral illustration and determines robotically the variety of change-points. Our detector comes with a proof of a non-asymptotic oracle inequality. Numerical experiments reveal the efficiency of the proposed methodology.