Some geometric surprises in modern machine learning
Andrea Montanari Stanford University
A flurry of recent work has studied the behavior of two-layers neural networks in the limit of a large number of neurons. I will review a subset of these ideas, outline open problems and present some new advances.