BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Contrastive Learning and High-Redshift Quasars - Xander Byrne 
DTSTART:20240221T140000Z
DTEND:20240221T150000Z
UID:TALK212530@talks.cam.ac.uk
CONTACT:Sri Aitken
DESCRIPTION:The basic task of unsupervised machine learning is to cluster 
 similar data together. Contrastive learning [Chen+20] has emerged as a pop
 ular clustering algorithm\, with the referenced paper garnering over a tho
 usand citations since the method's inception at Google at the start of thi
 s decade. It is based on a familiar concept ported from the field of super
 vised learning - data augmentation\, the expansion of a training set by ad
 ding innocuous transformations of training data\, such as rotations of an 
 image. Contrastive learning is fuelled by data augmentation\, teaching a n
 etwork to ignore such trivial changes\, improving generalisability whilst 
 it learns to cluster. This talk will give an overview of contrastive learn
 ing\, data augmentation\, and some examples where it has been used effecti
 vely\, including the discovery of three unique high-redshift quasars
LOCATION:Maxwell Centre
END:VEVENT
END:VCALENDAR