Spike and slab variational Bayes for high dimensional logistic regression

Neural Information Processing Systems 

Since the VB algorithm does not depend on the unknown truth to achieve optimality, our results shed light on effective prior choices.