Textbook in PDF format Algorithms that can improve their performance using training data Typically we have a large number of parameters Learned from data Useful if: No expert knowledge available: industrial/manufacturing control, mass spectrometer analysis, drug design, astronomic discovery Black-box expert knowledge: face/handwriting/speech recognition, driving a car, flying a plane Fast changing phenomena: credit scoring, financial modeling, diagnosis, fraud detection Customization/personalization: personalized news reader, movie/book recommendation Why do we see this “explosion” now? More training data More computation power New algorithms (deep learning, generative models)… – But: same principles as from 40 years ago are still in use "when you go from 10,000 training examples to 10 billion training examples, it all starts to work. Data trumps everything.”