Predicting the evolution of stationary graph signals

One way of tackling the dimensionality issues arising in the modeling of a multivariate process is to assume that the inherent data structure can be captured by a graph. We here focus on the problem of predicting the evolution of a process that is time and graph stationary, i.e., a time-varying signal whose first two statistical moments are invariant over time and correlated to a known graph topology. This stationarity assumption allows us to regularize the estimation problem, reducing the variance and computational complexity, two common issues plaguing high-dimensional vector autoregressive models. In addition, our method compares favorably to state-of-the-art graph and time-based methods: it outperforms previous graph causal models as well as a purely time-based method.