Resourceful Contextual Bandits

We study contextual bandits with ancillary constraints on resources, which are common in real-world applications such as choosing ads or dynamic pricing of items. We design the first algorithm for solving these problems, and prove a regret guarantee with near-optimal statistical properties.