Bayesian optimization is an approach to non-convex optimization that leverages a probabilistic model to make decisions about candidate points to evaluate. The primary advantage of this approach is the ability to incorporate prior knowledge about the objective function in an explicit way. While such prior information has typically been information about the smoothness of the function, many machine learning problems have additional structure that can be leveraged. I will talk about how such prior information can be found across tasks, within inner-loop optimizations, and in constraints.