## AbstractNowadays the rate and volume of information flow are sharply
increasing and forcing numerous domains to switch from offline to online
information processing. The tutorial will start with an overview of the classical online learning problems - prediction with expert advice, adversarial multiarmed bandits, stochastic multiarmed bandits, and bandits with side information (Cesa-Bianchi and Lugosi, 2006, Bubeck and Cesa-Bianchi, 2012). The Space of Online Learning ProblemsThen we will introduce Most of the classical results in online learning correspond to isolated points in the space of online learning problems. For example, prediction with expert advice, adversarial multiarmed bandits, and stochastic multiarmed bandits correspond to the three red crosses on the figure (in clockwise direction). In the second part of the tutorial we will describe a number of new algorithms that solve ranges of problems in the space of online learning problems (illustrated by dashed lines on the figure). The results include algorithms that interpolate between full information and bandit feedback (Mannor and Shamir, 2011, Alon et al., 2013, Seldin et al., 2014, Kale, 2014); algorithms that interpolate between adversarial and i.i.d. (or in some other sense “sub-adversarial”) environments in the full information (Gaillard et al., 2014, Koolen et al., 2014, Wintenberger, 2015) and in the bandit setting (Bubeck and Slivkins, 2012, Seldin and Slivkins, 2014); and algorithms that interpolate between stateless bandits and bandits with side information (Seldin et al.,2011). These results open a new era in online learning research, where the researchers progress from studying isolated points in the space of online learning problems to studying ranges of problems. ## Presenter
## Detailed ContentSee extended abstract. ## Slides |