Variational Inference for Complex Networks

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

Large-scale complex networks are becoming increasingly common in a variety of scientific disciplines, including social sciences, biological sciences, and physical sciences. Such complex networks challenge the computational limit of classical methods, making it infeasible to carry out statistical network inference within a reasonable amount of time. This project will develop efficient algorithms that are computationally feasible for large-scale complex networks and have provable statistical guarantees on performance. The proposed methods will be applied to social and biological network data, including brain networks, and will be used for the study of disorders associated with hearing loss, such as tinnitus. The proposed research is highly interdisciplinary and provides an opportunity for involvement of graduate and undergraduate students with a broad range of backgrounds and interests. The proposed methods will be incorporated into relevant courses. Research results will be disseminated to the scientific communities and all software developed in this research will be freely distributed as open-source to the public. The project will develop variational methods for complex networks, including dynamic, multi-layer, and heterogeneous networks, and investigate theoretical properties of the variational methods on these networks to provide provable statistical guarantees on performance. The network models the PI studies include latent space models for dynamic networks and dynamic multi-layer networks, stochastic block models for multi-layer networks, various models for heterogeneous networks, and other models for complex networks. The proposed variational inference procedure makes it possible to handle large scale complex network data. The theoretical properties the PI will investigate include consistency of parameter estimation and community detection for variational methods. The proposed methods will be applied to real network data from social and natural sciences. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.