Title: Fast and accurate variational inference in mixed models
Authors: Luca Maestrini - The Australian National University (Australia) [presenting]
Abstract: Mixed models with fixed and random effects are widely used to analyse longitudinal and multilevel data that can potentially be high dimensional and have a variety of measurement scales. In these complex settings, variational approximations may facilitate fast approximate inference for the parameters of mixed models. We explain how streamlined solutions to sparse matrix problems can be used for making fast variational Bayes inference for models with a high number of random effects, where Bayesian computation is typically hindered by the size of design matrices. Accuracy is also a crucial aspect in variational approximations, especially for models with non-Gaussian responses. We show that resampling methods can offer a valid remedy to the potential inaccuracy of variational approximations and illustrate the use of bootstrap for variational inference in mixed models.