No-Regret Learning in Games

Catalog description

Machine learning has become an indispensable part of many application areas, in science, engineering and business disciplines. However, in real-world applications, the environment is often not stationary as assumed by the standard ML framework. Instead, it consists of many agents, each of whom is involved in decision making processes and each of whose actions impact all others’ outcomes. The course has two components: online learning in adversarial environments where the game-theoretical component is embedded in the abstract adversarial environment; and multi-agent learning in which an underlying (unknown) game is driving the rewards/costs of all agents. We will spend 6 weeks on each. The course will provide students with a solid theoretical foundation learning in games and allow them to start accessing the expanding literature in related topics.