Date and Time: Friday, June 11, 2021, 3pm-4:30pm
Speaker: Constantinos Daskalakis (MIT)
Title: From von Neumann to Machine Learning: Equilibrium Computation and the Foundations of Deep Learning (John von Neumann Lecture)
Abstract: Deep Learning has enabled significant progress on single-agent learning challenges, much of that progress owed to the empirical success of gradient descent and its variants in computing local optima of non-convex optimization problems. In multi-agent learning problems, the role of single-objective optimization is played by min-max optimization or, more generally, equilibrium computation. Yet, the complexity of the latter in settings that are relevant to Deep Learning applications, such as GANs and multi-agent reinforcement learning, appears daunting. Gradient-descent based methods are often unstable, and the overall tractability of computing even local approximate equilibria has remained mysterious. We shed light on this challenge through a combination of optimization, complexity-theoretic, and topological methods, presenting obstacles and opportunities for Deep Learning in the multi-agent world.
Bio: Constantinos Daskalakis is a Professor of Computer Science at MIT, working on Computation Theory and its interface with Game Theory, Economics, Probability Theory, Machine Learning and Statistics. His work has resolved long-standing problems about the computational complexity of Nash equilibrium, and multi-item auctions, and now focuses on high-dimensional statistics and learning from biased, dependent, and strategic data. He has been honored with the Nevanlinna Prize by the International Mathematical Union as well as other awards including the ACM Doctoral Dissertation Award, the Kalai Prize from the Game Theory Society, the Sloan fellowship in Computer Science, the SIAM outstanding paper prize, the Simons investigator award, the ACM Grace Murray Hopper award, and the Bodossaki foundation distinguished young scientists award.
Event Organization: Zurich Center for Market Design
Introduction: Marina von Neumann Whitman