Abstract: Dynamic pricing is one of the most common examples of online decision problems. With the development of e-commerce and the massive real-time data in online platforms today, feature-based pricing has become increasingly important. Semi-parametric feedback structure is a natural formulation in such problems, and benefits from tools from non-parametric statistical estimation.

In this work, we study feature-based pricing with semi-parametric feedback structure. We propose a dynamic learning and decision algorithm that makes use of the classical idea of the tradeoff between exploration (statistical estimation) and exploitation (reward optimization). Under mild conditions, our proposed algorithm achieves near-optimal regret in terms of dependence on the time horizon. This result offers a new perspective on combining statistical learning and decision-making in the online decision context.