Near-Optimal Bayesian Active Learning at Scale: Submodular Surrogates and Beyond Yuxin Chen University of Chicago