Mathematics of Deep Learning Seminar: Florent Krzakala

Title: Teacher-student models: exactly solvable models for statistical learning

Abstract: The high-dimensional settings allow one to use powerful asymptotic methods from probability theory and statistical physics to obtain precise characterization of statistical learning problems. There is a decades-long tradition in statistical physics with building and solving such simple models, that are usually referred to as “teacher-student” models in this context. I will present and review examples of recent work in this direction, and will also try to attempt to answer the question: are these simplified models actually useful?