The Benefits of Learning with Strongly Convex Approximate Inference

We explore the benefits of strongly convex free energies in variational inference, providing both theoretical motivation and a new meta-algorithm. Using the duality between strong convexity and stabil