Consider using Gaussian process optimization as a learner

We should check out this cool project: Bayesian optimization. I suspect it secretly hides an N^3 fitting cost, but it's an interesting approach to consider.