**11****月22日 ****Distributed learning with big data**

**（曹锡华数学论坛）**

讲座题目：**Distributed learning with big data**

主讲人：Ding-Xuan Zhou教授

主持人：羊丹平 教授

开始时间：2015-11-2210:35-11:20

讲座地址：**中北校区理科大楼****A****座****A504****报告厅**

主办单位：数学系 科技处

报告人简介：Ding-Xuan Zhou is Professor at department of mathematics at City University of Hong Kong. He joined City University of Hong Kong as a Research Assistant Professor in 1996. His research interests include learning theory, wavelet analysis and approximation theory. He has published over 100 research papers and is serving on editorial board of the international journals Advances in Computational Mathematics, Analysis and Applications, Complex Analysis and Operator Theory and Journal of Computational Analysis and Applications.

He received a Joint Research Fund for Hong Kong and Macau Young Scholars from the National Science Fund for Distinguished Young Scholars in 2005 and a Humboldt Research Fellowship in 1993. He has co-organized over 10 international conferences and conducted more than 20 research grants.

He received his B.S., M.S., and Ph.D. in applied mathematics from Zhejiang University.

报告内容简介:

Distributed learning is a method to handle big data. It is based on adivide-and-conquer approach and consists of three steps: first we divide over-sized data into subsets and each data subset is distributed to one individual machine; then each machine processes the distributed data subset to produce one output; finally the outputs from individual machines are combined to generate an output of the distributed learning algorithm. It is expected that the distributed learning algorithm can perform as efficiently as one big machine which could process the whole oversized data, in addition to the advantages of reducing storage and computing costs. This talk demonstrates some mathematical analysis of distributed learning with least squares regularization schemes in reproducing kernel Hilbert spaces. We provide error bounds in the most general situation and present minimax learning rates for the distributed learning algorithm. Integral operators and their approximations play an important role in our study.