题  目： Inverse Scale Space: New Regularization Path for Sparse Regression

报告人： 严明 博士（UCLA）

时  间： 2013年12月13日    下午3:00―4:00

地  点： 东区管理科研楼 365英国上市官网1611会议室

Abstract: In modern real-world applications, it is not uncommon to have larger number of measured variables than the sample size for high-dimensional datasets. Conventional regression methods fail in these datasets, and sparse regression is needed. Sparse regression is also important in many other cases. The mostly used regularization path for sparse regression is minimizing an l1-regularization term. However this method and its variants have many disadvantages such as bias and introducing more irrelevant features. We introduce inverse scale space (ISS)-a new regularization path for sparse regression, which is unbiased and has better performance in selecting the relevant features. We show why ISS is better than l1-minimization theoretically, and the comparison of both methods is done on synthetic and real data. In addition, we developed an algorithm to accelerate ISS.

严明博士简介：

Dr. Ming Yan received the B.S. degree (2005) and the M.S. degree (2008) in computational mathematics from University of Science and Technology of China, Hefei, China, the Ph.D. degree (2012) in mathematics from University of California, Los Angeles. He worked as a postdoc fellow at Rice University, and he is a postdoc scholar in Department of Mathematics at UCLA.
His research covers a wide range of applied mathematical applications, including optimization and variational models, signal and image processing, compressive sensing, parallel and distributed computing, and sparse regression. He has published more than 10 peer-reviewed journal and conference papers. 

主办单位：  365英国上市官网

                 国家数学与交叉科学合肥分中心

欢迎广大师生参加！