Submodularity in Machine Learning and Vision