B0238
Title: Variational inference of Ising models and their applications
Authors: Tapabrata Maiti - Michigan State University (United States) [presenting]
Abstract: Ising models originated in statistical physics and have been widely used in modeling spatial data and computer vision problems. However, statistical inference of this model and its applications remain challenging due to the intractable nature of the normalizing constant in the likelihood. We propose a novel procedure for parameter estimation using variational Bayes (VB) under a pseudo-likelihood of a two-parameter Ising model, which is computationally efficient and outperforms the existing methods. Then, we extend the methodology to high-dimensional variable selection problems when the feature vector is structurally dependent. Dependent feature vector commonly occurs in genetics, neuroimaging, and image analysis. As time permits, we will also discuss some theoretical properties of the proposed procedures.
