In the quest for scalable Bayesian computational algorithms we need to exploit the full potential of existing methodologies. In this note we point out that message passing algorithms, which are very well developed for inference in graphical models, appear to be largely unexplored for scalable inference in Bayesian multilevel regression models. We show that nested multilevel regression models with Gaussian errors lend themselves very naturally to the combined use of belief propagation and MCMC. Specifically, the posterior distribution of the regression parameters conditionally on covariance hyperparameters is a high-dimensional Gaussian that can be sampled exactly (as well as marginalized) using belief propagation at a cost that scales linearly in the number of parameters and data. We derive an algorithm that works efficiently even for conditionally singular Gaussian distributions, e.g., when there are linear constraints between the parameters at different levels. We show that allowing for such non-invertible Gaussians is critical for belief propagation to be applicable to a large class of nested multilevel models. From a different perspective, the methodology proposed can be seen as a generalization of forward-backward algorithms for sampling to multilevel regressions with tree-structure graphical models, as opposed to single-branch trees used in classical Kalman filter contexts.
Submitted 20 Apr 2017 to Computation
Published 21 Apr 2017
Author comments: 8 pageshttp://arxiv.org/abs/1704.06064http://arxiv.org/pdf/1704.06064.pdf