B1573
Title: Continuously indexed graphical models
Authors: Kartik Waghmare - EPFL (Switzerland) [presenting]
Victor Panaretos - EPFL (Switzerland)
Abstract: Let X be a real-valued Gaussian process indexed by a set U. It can be thought of as an undirected graphical model with every random variable X(u) serving as a vertex. This graph is characterized in terms of the covariance of X through its reproducing kernel property. Unlike other characterizations in the literature, the characterization does not restrict the index set U to be finite or countable and, hence, can be used to model the intrinsic dependence structure of stochastic processes in continuous time/space. Consequently, the said characterization is not (and apparently cannot be) of the inverse-zero type. This poses novel challenges for the problem of recovery of the dependence structure from a sample of independent realizations of X, also known as structure estimation. A methodology is proposed that circumvents these issues by targeting the recovery of the underlying graph up to a finite resolution, which can be arbitrarily fine and is limited only by the available sample size. The recovery is shown to be consistent so long as the graph is sufficiently regular in an appropriate sense and convergence rates are provided. The methodology is illustrated by simulation and two data analyses.