Variational approximations for exponential random graph models

      Department of Systems Engineering and Engineering Management
                             The Chinese University of Hong Kong
Title:  Variational approximations for exponential random graph models 
Speaker: Lingjiong Zhu (Florida State University)
We study a model of strategic network formation with heterogeneous players, that converges to an exponential random graph model. The likelihood of observing a specific network is known up to an intractable normalizing constant, which is infeasible to compute. The standard estimation method uses an MCMC algorithm that generates the samples from the exponential random graph model to provide an estimate of the normalizing constant. We provide an alternative tractable method of estimation for a large class of exponential random graph models. We show that a mean-field variational approximation of the likelihood provides a lower bound for the normalizing constant, which becomes exact as the network size grows to infinity. We also provide exact bounds for the approximation error of the variational mean-field for fixed network size. Special cases including the model featuring extreme homophily will also be discussed. This is based on the joint work with Angelo Mele.
Lingjiong Zhu grew up in Shanghai and went to study in England, where he got BA from University of Cambridge in 2008. He then moved to the United States and received PhD from New York University in 2013. After a stint at Morgan Stanley, he went to work at University of Minnesota as Dunham Jackson Assistant Professor, before joining the faculty at Florida State University as an Assistant Professor in 2015.
This seminar is hosted by Prof. Xuefeng Gao.
Venue: Room 513,
      William M.W. Mong Engineering Building (ERB),
      (Engineering Building Complex Phase 2)
      The Chinese University of Hong Kong.
Thursday, May 19, 2016 - 08:30 to 09:30