EcoSta 2018: Registration
View Submission - EcoSta2018
A0350
Title: Gaussian variational approximation with a factor covariance structure Authors:  David Nott - National University of Singapore (Singapore) [presenting]
Victor Ong - National University of Singapore (Singapore)
Michael Smith - University of Melbourne (Australia)
Abstract: Variational approximations have the potential to scale Bayesian computations to large datasets and highly parameterized models. Gaussian approximations are popular, but can be computationally burdensome when an unrestricted covariance matrix is employed and the dimension of the model parameter is high. To circumvent this problem, we consider a factor covariance structure as a parsimonious representation. General stochastic gradient ascent methods are described for efficient implementation, with gradient estimates obtained using the so-called \textit{reparametrization trick}. The end result is a flexible and efficient approach to high-dimensional Gaussian variational approximation. We will illustrate the methodology for robust P-spline regression and some high-dimensional logistic regression models.