who: Carl Vondrick
when: 11/6 Wednesday 11:45a – 1:15p
where: Computer Science Building Rm 140
food: Athena’s Pizza
generous sponsor: ORACLE LABS
Learning from Unlabeled Video
Abstract:
I will discuss our research to use large amounts of unlabeled video in order to efficiently train models for visual recognition. Leveraging millions of videos, our work develops methods for machines to learn perception tasks such as anticipating human actions in the immediate future, tracking visual objects, and recognizing ambient sounds. We show how to take advantage of the natural context available in video in order to learn without human supervision, for example through the natural synchronization of vision and sound, or the temporal coherence of motion and color.
Bio:
Carl Vondrick is an Assistant Professor of Computer Science at Columbia University. Previously, he was a research scientist at Google. He completed his Ph.D. at the Massachusetts Institute of Technology in 2017.