Statistical Mechanics of Learning and Computation

Recent years have witnessed the successful application of time-honored techniques from the statistical physics of disordered systems, like the replica method and the cavity method, to understanding modern advances in machine learning and computation. We will develop the foundations of these methods, starting with a crash course in statistical mechanics, and then progressing to the basic theory of spin glasses, associative memories, random matrices, and random landscapes. We will additionally learn how to apply this theory to problems in learning and computation, including high dimensional statistics and deep learning. Overall, this foundations course will prepare students to read the growing interdisciplinary literature spanning physics, learning and computation.