Timezone: »

On Last-Iterate Convergence Beyond Zero-Sum Games
Ioannis Anagnostides · Ioannis Panageas · Gabriele Farina · Tuomas Sandholm

Thu Jul 21 03:00 PM -- 05:00 PM (PDT) @ Hall E #1208
Most existing results about last-iterate convergence of learning dynamics are limited to two-player zero-sum games, and only apply under rigid assumptions about what dynamics the players follow. In this paper we provide new results and techniques that apply to broader families of games and learning dynamics. First, we show that in a class of games that includes constant-sum polymatrix and strategically zero-sum games, the trajectories of dynamics such as optimistic mirror descent (OMD) exhibit a boundedness property, which holds even when players employ different algorithms and prediction mechanisms. This property enables us to obtain $O(1/\sqrt{T})$ rates and optimal $O(1)$ regret bounds. Our analysis also reveals a surprising property: OMD either reaches arbitrarily close to a Nash equilibrium or it outperforms the robust price of anarchy in efficiency. Moreover, for potential games we establish convergence to an $\epsilon$-equilibrium after $O(1/\epsilon^2)$ iterations for mirror descent under a broad class of regularizers, as well as optimal $O(1)$ regret bounds for OMD variants. Our framework also extends to near-potential games, and unifies known analyses for distributed learning in Fisher's market model. Finally, we analyze the convergence, efficiency, and robustness of optimistic gradient descent (OGD) in general-sum continuous games.

Author Information

Ioannis Anagnostides (Carnegie Mellon University)
Ioannis Panageas (UC Irvine)

I am an Assistant Professor in UC Irvine. Before that I was a MIT Postdoctoral Fellow working with Costis Daskalakis. I obtained my PhD in Algorithms, Combinatorics, and Optimization (ACO) at Georgia Tech, advised by Prasad Tetali. At Georgia Tech, I also obtained a MSc in Mathematics. I did my undergrad studies in National Technical University of Athens. I am interested in theory of computation and its interface with optimization, dynamical systems, probability and statistics, machine learning and their applications to game theory and multi-agent reinforcement learning.

Gabriele Farina (Carnegie Mellon University)
Tuomas Sandholm (Carnegie Mellon University)

Tuomas Sandholm is Angel Jordan Professor of Computer Science at Carnegie Mellon University. He is Founder and Director of the Electronic Marketplaces Laboratory. He has published over 450 papers. With his student Vince Conitzer, he initiated the study of automated mechanism design in 2001. In parallel with his academic career, he was Founder, Chairman, and CTO/Chief Scientist of CombineNet, Inc. from 1997 until its acquisition in 2010. During this period the company commercialized over 800 of the world's largest-scale generalized combinatorial multi-attribute auctions, with over $60 billion in total spend and over $6 billion in generated savings. He is Founder and CEO of Optimized Markets, Strategic Machine, and Strategy Robot. Also, his algorithms run the UNOS kidney exchange, which includes 69% of the transplant centers in the US. He has developed the leading algorithms for several general classes of game. The team that he leads is the two-time world champion in computer Heads-Up No-Limit Texas Hold’em poker, and Libratus became the first and only AI to beat top humans at that game. Among his many honors are the NSF Career Award, inaugural ACM Autonomous Agents Research Award, Sloan Fellowship, Carnegie Science Center Award for Excellence, Edelman Laureateship, Newell Award for Research Excellence, and Computers and Thought Award. He is Fellow of the ACM, AAAI, and INFORMS. He holds an honorary doctorate from the University of Zurich.

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors