Collapsed variational bayesian inference of the author-topic model: application to large-scale coordinate-based meta-analysis

摘要

The author-topic (AT) model has been recently used to discover the relationships between brain regions, cognitive components and behavioral tasks in the human brain. In this work, we propose a novel Collapsed Variational Bayesian (CVB) inference algorithm for the AT model. The proposed algorithm is compared with the Expectation-Maximization (EM) algorithm on the large-scale BrainMap database of brain activation coordinates and behavioral tasks. Experiments suggest that the proposed CVB algorithm produces parameter estimates with better generalization power than the EM algorithm.