Game-theoretic equilibrium concepts provide a sound definition of how rational agents should act in multiagent settings. To operationalize them, they must be accompanied by techniques to compute equilibria. In this talk I will describe some of my work on solving extensive-form games, a broad game class that can model sequential interaction, imperfect information, and outcome uncertainty. Practical equilibrium computation in extensive-form games relies on two complementary methods: abstraction techniques and equilibrium-finding algorithms based on first-order methods or regret minimization. For abstraction I developed the first general framework for analyzing solution quality when solving an abstraction. For equilibrium finding I developed a class of strongly-convex distance measures that can be combined with accelerated bilinear saddle-point solvers to achieve state-of-the-art convergence rate. Finally I will discuss my work on Stackelberg equilibrium, a solution concept that is popular e.g. in security applications. I developed a robust variant of this solution concept, which is important in practice where models may be misspecified.
Christian Kroer is a Ph.D. candidate in computer science at Carnegie Mellon University, where he is advised by Tuomas Sandholm. Christian’s research lies at the intersection of artificial intelligence, operations research, and economics. His thesis work is on solving large-scale sequential games through a combination of artificial intelligence and mathematical programming techniques. Christian also enjoys working on several related problems: facilitating large-scale electronic markets, data-driven approaches to understanding such markets, and robust modeling and optimization for games, markets, and machine learning. Christian received the 2016 Facebook Fellowship in economics and computation and was a runner-up in the INFORMS Computing Society student paper competition. He was an intern at Microsoft Research NYC and was intern and has been a consultant for Facebook Core Data Science. Christian's teaching interests include artificial intelligence, game theory, electronic markets, data science, and optimization.