EcoSta 2022: Start Registration
View Submission - EcoSta2022
A0499
Title: Learning cross-layer dependence structure for multilayer networks Authors:  Jonathan Stewart - Florida State University (United States) [presenting]
Abstract: Multilayer networks are a network data structure in which a set of nodes in a population of interest have multiple modes of interaction or relation (e.g., persons in a social network can have familial ties, friendships, professional relationships, and more). Each layer of the network corresponds to an individual network or graph defined through one mode of interactions or relations. We propose a 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. On the methodological side, we develop a class of models for both sparse and dense multilayer networks that focus on modeling cross-layer dependence and can incorporate node covariates. We elaborate algorithms for parameter estimation and model selection. On the theoretical side, our contributions include establishing non-asymptotic theoretical guarantees which establish rates of convergence in high-dimensional settings for both maximum likelihood estimators and maximum pseudo-likelihood estimators, as well as a non-asymptotic bound on the error of the multivariate normal approximation for our estimators. We demonstrate a method for model selection that controls the false discovery rate and show through simulations that our methods provide an accurate learning platform for learning the cross-layer dependence structure of multilayer networks.