Learning cross-layer dependence structure in multilayer networks
Multilayer networks are a network data structure in which elements in a population of interest have multiple modes of interaction or relation, represented by multiple networks called layers. We propose a novel class of models for cross-layer dependence in multilayer networks, aiming to learn how interactions in one or more layers may influence interactions in other layers of the multilayer network, by developing a class of network separable models which separate the network formation process from the layer formation process. In our framework, we are able to extend existing single layer network models to a multilayer network model with cross-layer dependence. We establish non-asymptotic bounds on the error of estimators and demonstrate rates of convergence for both maximum likelihood estimators and maximum pseudolikelihood estimators in scenarios of increasing parameter dimension. We additionally establish non-asymptotic error bounds on the multivariate normal approximation and elaborate a method for model selection which controls the false discovery rate. We conduct simulation studies which demonstrate that our framework and method work well in realistic settings which might be encountered in applications. Lastly, we illustrate the utility of our method through an application to the Lazega lawyers network.
READ FULL TEXT