Overview

In empirical game theoretic analysis (EGTA), the goal is to analyze or estimate various properties of a game, given only noisy blackbox (simulator) access to it.In particular, this means that we do not know the utility function of the game, and can only access samples of (per-player) utility values at any strategy profile (configuration of all player strategies) that, in expectation, match the utility of said profile.

Over the last several years, alongside my collaborators Amy Greenwald, and Enrique Areyan Viqueira, I have worked to bound the sample complexity of estimating equilibria and other properties of games in this setting. Furthermore, we have extended these ideas to mechanism design, wherein the goal is not just to estimate the properties of a game, but to modify a game or select its parameters in order to incentivize various types of strategic behavior and other desiderata.

Publications

• E. A. Viqueira, C. Cousins, A. Greenwald. Learning Competitive Equilibria in Noisy Combinatorial Markets, AAMAS 2021.
• E. A. Viqueira, C. Cousins, A. Greenwald. Improved Algorithms for Learning Equilibria in Simulation-Based Games, AAMAS 2020.
• E. A. Viqueira, A. Greenwald, C. Cousins, E. Upfal. Learning Simulation-Based Games from Data, AAMAS 2019.
• E. A. Viqueira, C. Cousins, Y. Mohammad, A. Greenwald. Empirical Mechanism Design: Designing Mechanisms from Data , UAI 2019.