Model fitting and Bayesian inference via power expectation propagation

We study a message passing approach to power expectation propagation for Bayesian model fitting and inference. Power expectation propagation is a class of variational approximations based on the notion of α-divergence that extends two notable approximations, namely mean field variational Bayes and expectation propagation. An illustration on a simple model allows to grasp benefits and complexities of this methodology and sets the basis for applications on more complex models