New Paradigms for Exploiting Parallel Experiments in Bayesian Optimization