Deep Learning
A subfield of machine learning with models of multiple processing layers that learn their own features, at increasing levels of abstraction, instead of the human designing them.
When to use
Use it as the decision framework from Lecture 8: always start with a traditional baseline, and consider deep learning only with plenty of data, complex feature interactions and tolerance for a black box. On Derech's data the MLP reached 83% accuracy versus 72% for the best earlier model.
See Lec 8.