October 13, 2020

12:00 pm / 1:15 pm

Venue

ZOOM

Recorded Seminar: https://wse.zoom.us/rec/share/CTh7p9TYN2xvpxpK6M42xOs8EFgdYA2pn9oWoIdTrzlTTRgLSh830_1qIqMhg_0.8S83xUpa39xJTXO-?startTime=1602604189000:  https://wse.zoom.us/rec/share/CTh7p9TYN2xvpxpK6M42xOs8EFgdYA2pn9oWoIdTrzlTTRgLSh830_1qIqMhg_0.8S83xUpa39xJTXO-?startTime=1602604189000
Kate Saenko, PhD
Associate Professor
Computer Science
Boston University
and Consulting Professor
MIT-IBM Watson AI Lab

“Learning from Small and Biased Datasets?    

Abstract:DeepLearning has made exciting progress on many computer vision problems such asobject recognition in images and video. However, it has relied on largedatasets that can be expensive and time-consuming to collect and label.Datasets can also suffer from ?dataset bias,? which happens when the trainingdata is not representative of the future deployment domain. Dataset bias is amajor problem in computer vision — even the most powerful deep neural networksfail togeneralize to out-of-sample data. A classic example of this is when anetwork trained to classify handwritten digits fails to recognize typed digits,but this problem happens in many situations, such as new geographic locations,changing demographics, and simulation-to-real learning. Can we solve datasetbias and learn with only a limited amount of supervision? Indeed, we can, undercertain assumptions. I will describe some recent work based ondomainadaptation of deep learning models and point out several assumptionsthey makeand situations they fail to handle. I will also describe recent efforts toimprove adaptation by using unlabeled data to learn better features, with ideasfrom self-supervised learning.

Bio:  Kate is an Associate Professor ofComputer Science at Boston University and a consulting professor for theMIT-IBM Watson AI Lab. She leads the Computer Vision and Learning Group at BU,is the founder and co-director of the Artificial Intelligence Research (AIR)initiative, and member of the Image and Video Computing research group. Katereceived a PhD from MIT and did her postdoctoral training at UC Berkeley andHarvard. Her research interests are in thebroad area of ArtificialIntelligence with a focus on dataset bias, adaptive machine learning, learningfor image and language understanding, and deep learning.