From Points to Probability Measures: Statistical Learning on Distributions with Kernel Mean Embedding