Towards Lifelong Learning from Images and Text

Presenter: Svetlana Lazebnik, Department of Computer Science, University of Illinois

Tuesday, December 7, 2021

Slides: https://uofi.app.box.com/s/rv9nal8cul6u7th3utl9pm87sb86bwtq

Video: https://uofi.app.box.com/s/qgrsnxtc6b2u4b2w1g543w8nz90721d6

Abstract

In this talk, I will discuss my group’s work on cross-modal embeddings, visual grounding, and the applications of these techniques to vision-language understanding tasks. I will also discuss our work on multi-task and incremental learning. Finally, I will briefly survey the state of the art in the area of lifelong learning from images and text, and talk about key challenges and directions for advancing this field further.

 

Biography

Svetlana Lazebnik received her Doctor of Philosophy degree in Computer Science at University of Illinois in 2006. After serving as assistant professor at the University of North Carolina at Chapel Hill from 2007 to 2011, she returned as faculty the University of Illinois, where she is currently Full Professor in the Department of Computer Science. Lazebniks notable awards include the 2013 Sloan Research Fellowship, the 2009 Microsoft Research Faculty Fellowship, and the 2008 NSF CAREER Award. Her Ph.D. work on Spatial Pyramid Models for object recognition, published in CVPR 2006, received the 2016 Longuet-Higgins Prize for a paper with significant impact on computer vision. She was elected IEEE Fellow in 2021. Her main research themes include scene understanding, modeling of large-scale photo collections, joint representations for images and text, and deep learning techniques for visual recognition problems.