Since no conclusive evidence for the nature of physics beyond the Standard Model has yet been found, it makes sense to add new analysis methods to our arsenal. By using machine learning to interface with current analysis methods, we may be able to make significant gains in discovery sensitivities for a broad range of new physics models. This talk will focus on weakly supervised anomaly detection methods, which frame the anomaly detection task as a classification. Classification methods are fundamental to machine learning and exceptionally well studied. As such, we can draw on the vast experience of the computer science community to improve our methods, while at the same time drawing on the physics intuition of our community to improve data handling and the interface between machine learning and analysis as a whole.