Below is a talk I gave at the MIT Broad Institute on how understanding new results on submodular functions can lead to better algorithms for machine learning and statistics: