Efficient and Scalable Machine Learning Methods for Robust Bayesian Optimal Experimental Design