Problems of sequential decision-making under partial information commonly arise in the design of algorithms for networked systems and the applications they support.  In such tasks, a decision-maker must repeatedly choose from a set of alternatives, given only partial knowledge of the past costs or benefits of these alternatives and no knowledge of the future. Classically, such problems have been modeled as multi-armed bandit problems, and they have been extensively studied and applied in a broad range of contexts including machine learning theory, economics, game theory, and the design of experiments. In this talk, Robert Kleinberg presents two such problems, motivated by applications to routing in overlay networks and pricing in e-commerce. From the Series:CSE Colloquia - 2005