A #MachineLearning view of Distribution Estimation : Efficient Computation of Empirical Proper Losses , Mixed in/out-of-sample Asymptotics of Empirical Proper Losses , a Unified #MachineLearning Interface for Density Estimation , and a Systematic Benchmarking Experiment - @UCL Discovery