Title: Bayesian inference in large complex networks
Authors: Snigdhansu Chatterjee - University Of Minnesota (United States) [presenting]
Abstract: Datasets are considered where the observations are over vertices of large graphs or networks, and there may be high-dimensional features associated with each vertex. For such datasets, the nature of the observations may depend not only on the features of the vertex they are associated with, but also on features of other vertices depending on the properties of the edges. We discuss Bayesian inference in models for such data. We consider generalizations of traditional network models like the stochastic block model, random dot product model, and so on, and propose different computational strategies for obtaining posterior distributions of interest. Apart from the traditional Markov Chain Monte Carlo approach, we study approximate Bayesian computations and some recently proposed piecewise deterministic Monte Carlo approaches. Comparisons between the different algorithms based on their computational efficiencies and statistical accuracies are discussed. Applications in several real data problems are discussed.