great talk, I'd only add that CS schools don't teach ML algorithms, they teach theoretical models. Just look at Bishop's PRML book, which is the de facto standa

great talk, I'd only add that CS schools don't teach ML algorithms, they teach theoretical models. Just look at Bishop's PRML book, which is the de facto standard textbook these days or Andrew Ng's lectures on Youtube. While they present a good overview of Learning Theory Concepts, the actual algorithms are largely left to students to devise (or use existing ML libraries in black-box mode). The distance between academia and industry is even larger than what Ilya describes.