Design of algorithms for games and learning

This proposal lies at the intersection of learning theory, game theory, and mechanism design and plans to investigate the connections and expand the mathematical foundations of these fields.

During the last decades, the research community intensified its efforts to understand systems with strategic participants and develop methods to influence and, if possible, to control them, employing game theory and mechanism design. The main drive has been the rapid growth of the Internet, the Web, and recently, social media and blockchains, systems that involve a variety of autonomous entities with different and often conflicting interests. At the same time, machine learning has made tremendous progress in using sets of data for generalization and prediction with a wide spectrum of applications.WP1 covers settings in which the predictive power of machine learning can be used to improve known algorithmic limitations of mechanism design. WP2 has the opposite flavor in that it uses mechanism design techniques to deal with incentives, in machine learning settings where the input is generated by selfish parties. Finally, WP3 considers traffic systems, which have always been affected by the drivers’ strategic considerations, but with the expansion of self-driving cars these considerations become more prevalent and transparent. We will employ both machine learning and game theory techniques to design efficient traffic networks.