Contextual Bandits with Continuous Actions: Smoothing, Zooming, and Adapting
conference on learning theory, pp. 2025-2027, 2019.
We study contextual bandit learning with an abstract policy class and continuous action space. We obtain two qualitatively different regret bounds: one competes with a smoothed version of the policy class under no continuity assumptions, while the other requires standard Lipschitz assumptions. Both bounds exhibit data-dependent "zooming...More
PPT (Upload PPT)