Title: Learning with Deep Generative Models

Abstract: Deep generative models are flexible tools on revealing the latent structures underlying complex data and performing “top-down” inference to generate samples. In this talk, I will present some results on learning with deep generative models, including discriminative learning for high-accuracy prediction in supervised and semi-supervised settings, a new conditional moment-matching criterion to estimate the parameters, and ZhuSuan--a GPU library to support probabilistic programming and efficient inference. 

朱军，清华大学计算机系长聘副教授、卡内基梅隆大学兼职教授，智能技术与系统国家重点实验室副主任。担任顶级期刊IEEE PAMI和Artificial Intelligence编委，顶级会议ICML 2014地区联合主席, ICML (2014-2017)、NIPS (2013, 2015)、IJCAI（2015、2017）、AAAI（2016, 2017）等领域主席。获CCF青年科学家奖、国家优青基金、中创软件人才奖等，入选国家“万人计划”青年拔尖人才计划和IEEE Intelligent Systems AI’s 10 to Watch。