组织者 Organizer：周源

Abstract: We study the PAC learnability of revenue maximization for a single buyer with an unknown value distribution, under both sample access and adaptive pricing-query access. The goal is to post a price that achieves a $(1 - \varepsilon)$-multiplicative approximation to the optimal revenue with probability $(1 - \delta)$.

Up to constant factors, we establish a tight characterization of the sample complexity, the (single-sample) query complexity, and the full query-sample tradeoff for the canonical families of \textit{$\lambda$-regular} and \textit{$\lambda$-quasi-regular} distributions ($0 \le \lambda \le 1$), including the four endpoint families: \textit{monotone-hazard-rate (MHR)}, \textit{regular}, \textit{quasi-monotone-hazard-rate (quasi-MHR)}, and \textit{quasi-regular} \cite{BMP63, M81, CR14, HR14, FJ25}. Our results tighten the known sample- and query-complexity bounds and, through the new perspective on the query-sample tradeoff, bridge the two previously parallel lines of research.

Bio: Yaonan Jin is an Assistant Professor in the Department of Computer Science and Engineering at the Hong Kong University of Science and Technology. Before joining HKUST, he conducted theoretical computer science research at Huawei's Taylor Lab, working with Pinyan Lu. He obtained his PhD from Columbia University in 2023 (advised by Xi Chen and Rocco Servedio). Prior to that, he obtained his MPhil from Hong Kong University of Science and Technology (advised by Qi Qi) and his BEng from Shanghai Jiao Tong University.