The reasonable power of the KL divergence
From intuition to deterministic inference
When doing Bayesian Inference, or any other type of inference, chances are you have heard about KL divergence – short for Kullback-Leibler divergence. This quantity has been pervasive in machine learning and artificial intelligence and in this post I would like to explore with you, the reader, some reasons for why this is so, especially for probabilistic inference. Eventually, we’ll find ourselves dealing with Variational Inference and with a little more patience, with Expectation Propagation.
Read more...