Nearly Minimax-Optimal Regret for Linear Contextual Bandits with Arbitrary Adaptive Action Sets
Organizations: Data Science and Analytics Thrust The Hong Kong University of Science and Technology (Guangzhou)
Abstract
We study stochastic linear contextual bandits with arbitrary action menus that may depend on the fixed parameter and the interaction history. We establish matching upper and lower bounds, up to logarithmic factors. Let be the dimension, be the menu size, and the time horizon. For , we prove an upper bound . When , we further prove a lower bound . Thus, for and , the upper and lower bounds match up to logarithmic factors, and the polynomial dependence on is optimal. Compared with the previous bound, our upper bound improves the dependence on by a factor of . For , we prove an upper bound and a lower bound . Here, omits logarithmic factors only in and . In particular, for polynomially large , the upper and lower bounds both scale as up to logarithmic factors, improving the standard rate by a factor of . As grows further, the regret smoothly recovers the scale once reaches order .