BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:CAMS Colloquium: A.I. and Its Implications for Mathematical Re
 search: Understanding and Characterizing Regularization: Learnability and 
 Physics-Based Energy Guidance
X-WR-TIMEZONE:Pacific Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260718T024526Z
UID:tag:localist.com\,2008:EventInstance_51376364380162
DTSTART:20260309T223000Z
DTEND:20260309T233000Z
DESCRIPTION:Shanghua Teng\, USC\n\n\nTitle: Understanding and Characterizin
 g Regularization: Learnability and Physics-Based Energy Guidance\n\n\nAbst
 ract: The quintessential learning algorithm of empirical risk minimization
  (ERM) is known to fail in various settings for which uniform convergence 
 does not characterize learning. Relatedly\, the practice of machine learni
 ng is rife with considerably richer algorithmic techniques\, perhaps the m
 ost notable of which is regularization. Nevertheless\, no such technique o
 r principle has broken away from the pack to characterize optimal learning
  in these more general settings. The purpose of this research direction is
  to understand the role of regularization in data-driven machine learning.
 \n\nFirst\, we focus on an abstract statistical learning framework\, prese
 nt our work on characterizing the power of regularization in perhaps the s
 implest setting for which ERM fails: multiclass learning with arbitrary la
 bel sets. Using one-inclusion graphs (OIGs)\, we exhibit a local-regulariz
 ation approach to obtain optimal learning algorithms that dovetail with tr
 ied-and-true algorithmic principles: Occam’s Razor as embodied by struct
 ural risk minimization (SRM)\, the principle of maximum entropy\, and Baye
 sian inference.\n\n\nSecond\, we share some of our on-going progress on de
 signing physics-guided energy regularization for data-driven learning of t
 he weak solution to parabolic PDEs. The goal here is to extend semi-superv
 ised learning by exploiting auxiliary data and the underlying physical mod
 el to construct stronger regularization\, enabling more efficient learning
  with optimal estimation and faster generalization.\n\n\nJoint work (COLT 
 2024) with Julian Asilis\, Siddartha Devic\, Shaddin Dughmi\, and Vatsal S
 haran\n\n\nJoint work with Xiaohui Chen and Zixiang Zhou.
GEO:34.022409;-118.291027
LOCATION:Kaprielian Hall (KAP)\, 414
SUMMARY:CAMS Colloquium: A.I. and Its Implications for Mathematical Researc
 h: Understanding and Characterizing Regularization: Learnability and Physi
 cs-Based Energy Guidance
URL;VALUE=URI:https://calendar.usc.edu/event/cams-colloquium-ai-and-its-imp
 lications-for-mathematical-research-8029
CATEGORIES:Lecture / Talk / Workshop
END:VEVENT
END:VCALENDAR