From Shape-Constrained Regression to Feature Detection

时间:2022-01-04         阅读:


主题From Shape-Constrained Regression to Feature Detection

主讲人伦敦政治经济学院 陈一宁助理教授

主持人统计学院 常晋源教授


举办地点:腾讯会议,244 774 905

主办单位:数据科学与商业智能联合实验室 统计学院 科研处


Yining Chen is an assistant professor in statistics at the London School of Economics; he did his PhD at the University of Cambridge; his current research interests include non-parametric (especially shape-constrained) problems, change-point detection and statistical computing.



In this talk, I will first revisit some of the commonly-used tools in shape-constrained regression problems, where shapes we consider including isotonic, unimodal, convex/concave regression. I will also talk about more recent work on S-shape regression, where the regression function consists of both convex and concave parts. In the second part, I shall discuss how these estimators provide natural candidates for estimating the location of features of the regression function, such as the mode and the inflection point. Finally, I will discuss how these tools can be used together with techniques from the change-point detection literature to detect useful (potentially multiple) features of a regression function. (This talk is based on joint work with Oliver Feng, Qiyang Han, Ray Carroll, Richard Samworth and Piotr Fryzlewicz.)

本报告将首先回顾一些在形状约束回归问题中常用的工具,包括等渗、单峰、凸/凹回归。然后将讨论S形回归的最新工作,其中的回归函数是由凸部分和凹部分组成。在第二部分,将讨论这些估计量如何为估计回归函数特征的位置提供自然候选,例如众数和拐点。最后,将讨论如何将这些工具与变化点检测文献中的技术结合使用,以检测回归函数的有用(可能多个)特征。(本报告基于与Oliver Feng、Qiyang Han、Ray Carroll、Richard Samworth和Piotr Fryzlewicz的合作。)

    <label id='TMha'><person></person></label><person id='yNDZ'><small></small></person><tt id='OrLv'><dir></dir></tt><dir id='QcKPNVI'><caption></caption></dir>
      <basefont id='rDyvwMQ'><ol></ol></basefont>
      <dir id='sphgL'><comment></comment></dir>
          <listing id='ii'><span></span></listing>
            <em id='gELd'><ins></ins></em>