A Full Classification of Finite Adversarial Partial Monitoring - Tor Lattimore
Explore a comprehensive 42-minute lecture on the full classification of finite adversarial partial monitoring, presented by Tor Lattimore at the Alan Turing Institute. Delve into the theoretical foundations of learning, focusing on methods that intersect statistics, probability, and optimization. Examine how partial monitoring generalizes the multi-armed bandit framework, encompassing bandit games, full-information games, and variants in between. Discover the core question of how game structure influences regret dependence. Learn about the complete classification of finite adversarial partial monitoring, which categorizes all games into four distinct categories. Gain insights from joint research with Csaba Szepesvari, as presented in their arXiv paper.