题目：A Scalable L0-Based Sparse Regression Method for Large-Scale Competing Risks Data

摘   要

In this talk I will present a scalable L0-based simultaneous variable selection and parameter estimation method for the popular Fine-Gray (1999) proportional subdistribution hazards (PSH) model for large competing risks time-to-event data via the broken adaptive ridge (BAR) method. We first establish that the BAR estimator, defined as the limit of an L0-based iteratively reweighted L2-regularization algorithm, is selection consistent and has an oracle property for the PSH model. To make it scalable to large studies, we further develop two novel high performance computational algorithms: 1) a cyclic coordinate-wise BAR (cycBAR) algorithm that effectively eliminates the need to fitting multiple reweighted L2-regularizations, and 2) a forward-backward scan algorithm that reduces the computation costs for the PSH model from O(n^2) to O(n). In comparison to standard methods, our new algorithms can produce over 1,000-fold speedups in computation time for the PSH model. Illustrations of the impressive scalability of our methods for large competing risks data is given using both simulations and real data.

   报告人：Li, Gang教授  加州大学洛杉矶分校生物统计系

报告人简介：Li, Gang教授是国际数理统计学会、美国统计学会、国际统计、皇家统计学会会士，发表学术论文120多篇。担任Jonsson综合癌症中心BASE Unit Shared Resource、加州大学洛杉矶分校Center of Excellence in Pancreatic Cancer主任。

  时间：2019年6月10日（星期一） 上午10:10-11:10

地点：数学科学学院424室

欢迎广大师生参加!

数学科学学院

 2019年6月6日