Notice: This lecture is a research seminar for credit.

Time

————————————

Mar 27, 2019 9:00-10:00 a.m.

Abstract

————————————

Data-driven applications  are often confronted with many practicalchallenges, from uncertainty in the collected data to the lack of dataannotations for complicated tasks. Fortunately, large data  collections often exhibit regular structures,which essentially represent the  “sharedknowledge” among similar data instances or tasks.   Exploiting such structures is often the keyin designing robust, scalable data-driven algorithms.

In this talk, I will presentfindings from my past research on extracting, utilizing and understandingshared knowledge  in designingdata-driven algorithms. These findings show how shared structures can reduce  the uncertainty in the input data, and makelearning new tasks more efficient and cost effective.  I will highlight several applications ofshared knowledge in GPS trajectory data under uncertainty and a metric toquantify the effectiveness of shared knowledge in task transfer learning.Additionally, I will discuss the open question of finding betterrepresentations of shared knowledge in a multi-task scenario.

Speaker's Bio

————————————

Yang Li received a B.A. degree in2011, from the Department of Computer Science, Smith College, and a Ph. D.degree, in 2017, from the Department of Computer Science, Stanford University.Since 2017, she has been working as a postdoc at  the Internet of Things and Societal CyberPhysical System Lab (2C)  of TBSI.  Her research interests include computationalgeometry, spatial algorithms, data mining and machine learning.

Registration

————————————

Professors and students of TBSI are welcometo attend. The lecture is also open to the public. For off-campuspersonnel, please scan the QR code and and fill in your information (name,company, contact number, ID number). The language of the lecture isEnglish.