Abstract：Classical statistical methods including local polynomial regression, random forests and many others can be regarded as weighted average type estimation based on specific kernels. We discuss how to enhance semi-supervised learning and deep learning incombination with suitable kernel designs. We also provide theoretcial guarantees for our method. We also present a multitude of numerical studies to demonstrate the prediction power of our proposal.

 

Bio: 於州，华东师范大学教授、博士生导师。主要研究方向为高维数据统计分析及统计机器学习，在Annals of Statistics, Biometrika，JASA, JRSSB, Journal of Machine Learning Research, IEEE Information Theory等知名统计及机器学习期刊上发表论文50余篇。曾主持国家重点研发计划课题、自然科学基金青年、面上项目，获得上海市自然科学二等奖等奖项，霍英东青年科学奖二等奖。并先后入选上海高校东方学者特聘教授，上海市青年拔尖人才，国家级青年人才等计划。