讲座题目：Variance Reduced Median-of-Means Estimator for Byzantine-RobustDistributed Inference

主讲人：刘卫东教授 上海交通大学

讲座时间：2020年9月1日（周二）下午2：00-4：00

参与方式： 本场报告将通过腾讯会议举办，

会议 ID：508382 648

会议直播: https://meeting.tencent.com/l/C7OlU75b7abf

主讲人简介：

刘卫东，上海交通大学数学科学学院副院长，特聘教授，国家级高层次青年人才。2008年于浙江大学获博士学位，2008-2011年在香港科技大学、美国宾夕法尼亚大学沃顿商学院从事博士后研究工作。2010年获全国百篇优秀博士学位论文奖及由世界华人数学家大会颁发的新世界数学奖；2013年获得国家级青年人才基金资助；2016年获得国家级高层次领军人才称号；2018年获国家级高层次青年人才基金资助。研究兴趣包括现代统计学、机器学习等，在统计学四大顶级期刊（AOS,JASA,JRSSB,Biometrika）和机器学习顶级期刊JMLR发表40余篇论文。

讲座。

讲座摘要：

This paper develops an efficientdistributed inference algorithm,  which is robust against a moderatefraction of Byzantine nodes, namely arbitrary and possibly adversarial machinesin a distributed learning system. In robust statistics, the median-of-means(MOM) has been a popular approach to hedge against Byzantine failures due toits ease of implementation and computational efficiency. However, the MOMestimator has the shortcoming in terms of statistical efficiency. The firstmain contribution of the paper is to propose a variance reduced median-of-means(VRMOM) estimator, which improves the statistical efficiency over the vanillaMOM estimator and is computationally as efficient as the MOM. Based on theproposed VRMOM estimator, we develop a general distributed inference algorithmthat is robust against Byzantine failures.  Theoretically, our distributedalgorithm achieves a fast convergence rate with only a constant number ofrounds of communications. We also provide the asymptotic normality result forthe purpose of statistical inference. To the best of our knowledge, this is thefirst normality result in the setting of Byzantine-robust distributedlearning.  The simulation results are also presented to illustratethe effectiveness of our method.