Entropy Search (ES) and Predictive Entropy Search (PES) are popular and
empirically successful bayesian optimization techniques. Both rely on a
compelling information-theoretic motivation, and maximize the information
gained about the $\arg\max$ of the unknown function; yet, both are p