Online recovery of time-varying signals defined over dynamic graphs

The goal of this work is to devise least mean square (LMS) strategies for online recovery of time-varying signals defined over dynamic graphs, which are observed over a (randomly) time-varying subset of vertices. We also derive a mean-square analysis illustrating the effect of graph variations and sampling on the reconstruction performance. Finally, an optimization strategy is developed in order to design the sampling probability at each node in the graph, with the aim of finding the best tradeoff between steady-state performance, graph sampling rate, and learning rate of the proposed method. Numerical simulations carried out over both synthetic and real data illustrate the good performance of the proposed learning strategies.