The Science of Modern Machine Learning: New Opportunities for Theory

Deep learning, the technology underlying the recent progress in AI, has revealed some major surprises from the perspective of theory. These methods seem to achieve their outstanding performance through different mechanisms from those of classical learning theory, mathematical statistics, and optimization theory. Simple gradient methods find excellent solutions to nonconvex optimization problems, and without any explicit effort to control model complexity, they exhibit excellent prediction performance in practice. This talk will describe recent progress on the optimization and generalization properties of these methods, as well as some of the intriguing questions that they raise.

Speaker: Peter Bartlett, UC Berkeley and Google DeepMind