•   When: Tuesday, April 04, 2017 from 02:30 PM to 03:30 PM
•   Speakers: Shandian Zhe
•   Location: Research Hall, Room 163
•   Export to iCal

Abstract:

Discovering and understanding complex, hidden relationship in real-world data is a key problem in machine learning and data mining. For example, modelling the interactions between users and items in online transactions is critical for accurate commodity recommendation; discovering the correlated relationship between genes and cancers can provide important insights in helping disease diagnosis and medicine development.

 Bayesian learning provides a highly principled and interpretable mathematical framework for data modeling and reasoning under uncertainty. In this talk, I will discuss my works on Bayesian learning techniques to uncover the hidden, complex relationship in real-world large data. In the first part, I will introduce my works on nonlinear tensor factorization, where I use Bayesian nonparametric models to capture the nonlinear interactive relationship underlying tensor data and develop parallel inference algorithms on Hadoop and Spark platforms. I demonstrate their impressive accuracy gains for tensor completion tasks in billion-entry level data. In the second part, I will introduce my works of Bayesian sparse learning in extracting correlated relationship between features and responses. I show their successful applications in Alzheimer’s disease studies and large-scale click-through-rate prediction tasks with millions of samples and hundreds of thousands of features.

Bio

 Shandian Zhe is a Ph.D. candidate in Department of Computer Science at Purdue University. He was awarded the Google Ph.D. Fellowship for machine learning in 2016. His research contributions have been published at NIPS, AISTATS, KDD, AAAI and IJCAI. He was nominated for the outstanding student paper in AAAI 2015. His primary research interests lie in Bayesian machine learning and data mining.