BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//NextToU//EN
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:dfe218c1-8b11-435b-8f96-63abd87966f5@nexttoyou
DTSTAMP:20261105T155928Z
DTSTART:20261109T190000Z
DTEND:20261109T200000Z
SUMMARY:ECE Seminar Series: Vishwanath Saragadam
LOCATION:Winston Chung Hall 205/206
DESCRIPTION:Flyer: Beyond Large-Scale RGB: Harnessing Priors and Physics A
 cross Data-Scarce Regimes\n\nAbstract: The recent revolution in computer 
 vision has largely been propelled by the availability of massive\, annota
 ted RGB datasets. However\, in domains such as infrared and other wavelen
 gths\, such large-scale labeled data is virtually nonexistent. As the amo
 unt of data shrinks\, purely data-driven paradigms break down\, necessita
 ting a shift toward strong domain priors and physical models. In this tal
 k\, I will explore three distinct regimes to address these challenges. Fi
 rst\, when no training data is available\, we rely on the implicit biases
  of neural networks to perform tasks directly from single observations. S
 econd\, when only a small amount of data is available\, we leverage natur
 al image statistics to regularize learning and improve generalization. Th
 ird\, when models are well trained but noise is correlated\, we incorpora
 te the physics of image formation to restore robustness and accuracy. Tog
 ether\, these strategies provide a framework for expanding computer visio
 n beyond the limits of RGB by bridging data scarcity through physics and 
 structural priors.\n\nhttp://next-to-u.app/events/dfe218c1-8b11-435b-8f96
 -63abd87966f5
URL:http://next-to-u.app/events/dfe218c1-8b11-435b-8f96-63abd87966f5
END:VEVENT
END:VCALENDAR