Learning Graph ARMA Processes from Time-Vertex Spectra

ANY modern digital platforms involve the acquisition of data over networks, while network data has a stationary process models, ARMA models widely used in typically time-varying structure. For instance, measurements classical signal processing have also been adapted to graph acquired on a sensor network or user data in a social network domains in several recent works [5], [6]. Meanwhile, the often vary over time. Such data can be modeled as timevarying computation of an ARMA process model is a challenging graph signals, or time-vertex signals. In many practical problem in graph domains as it typically involves the solution applications, time-vertex signals may have missing observations of highly nonlinear and nonconvex optimization problems. The due to issues such as sensor failure, connection loss, and problem of learning graph ARMA process models has been partial availability of user statistics. Hence, the spatio-temporal addressed in the previous studies [3], [5], [6]; however, none of interpolation of time-vertex signals arises as an important these studies explicitly aim to capture the specific time-vertex problem of interest.