Learning Representations from Data
朱军
Professor
清华大学计算机科学与技术系
Abstract: As the availability and scope of complex data increase in both scientific and engineering fields, developing statistical tools to discover latent structures and hidden explanatory factors has become a major theme of statistic and machine learning research. Breakthrough work has recently been done on learning latent feature representations with shallow or deep architectures using a huge amount of computing resources and massive data corpora. However, some key research issues are still remaining under-addressed. In this talk, I will introduce some of our recent work on learning representations that are discriminative in specific application tasks, including classification, multi-view data analysis, social link prediction, and low-rank matrix factorization. I will share some insights on dealing with several key issues on learning latent representations, including discriminative ability, model complexity, sparsity/interpretability, and scalability.
Biography: 朱军,清华大学计算机科学与技术系副教授、博士生导师,中国计算机学会优秀论文奖、微软学者、以及国家优秀青年基金获得者,入选清华大学221基础研究人才计划。2009到2011年在美国卡内基梅隆大学机器学习系做博士后研究。主要从事机器学习、贝叶斯统计等基础理论、算法及相关应用研究。相关工作在国际期刊与会议JMLR, PAMI, ICML, NIPS等发表论文40余篇。受邀担任机器学习顶级国际会议ICML2014的联合地区主席、ICML2014和NIPS2013的领域主席、IJCAI2013的资深程序委员等。2013年入选IEEE Intelligent Systems国际杂志评选的 “AI’s 10 to Watch”。