BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:ISE Seminar Series 
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260616T191150Z
UID:tag:localist.com\,2008:EventInstance_52638002479458
DTSTART:20260421T150000Z
DTEND:20260421T160000Z
DESCRIPTION:Guest Speaker: Dr. Lin Xiao\, Meta\n\nLecture: Quantization thr
 ough Piecewise-Affine Regularization\n\nAbstract: Quantization is a widely
  adopted approach for model compression\, which can significantly reduce t
 he memory\, compute\, and latency of modern AI models. However\, current q
 uantization methods are mostly based on heuristics and lack theoretical su
 pport. We propose a general framework of piecewise-affine regularization (
 PAR) for finding quantized solutions to continuous optimization problems a
 nd investigate its statistical properties and optimization methods. \n\nWe
  show that in the over-parameterized regime of supervised learning\, every
  critical point of the PAR-regularized loss function exhibits a high degre
 e of quantization. For linear regression problems\, we can approximate the
  classical formulations of ridge regression and Lasso using PAR and obtain
  similar statistical guarantees with quantized solutions. For optimization
  methods\, we focus on the setting with a stochastic gradient oracle and d
 iscuss algorithms that can successfully identify the manifold of quantized
  solutions. \n\nBio: Lin Xiao is a Research Scientist at Meta's Fundamenta
 l AI Research (FAIR) lab. He received a Bachelor of Engineering from Beiji
 ng University of Aeronautics and Astronautics (now Beihang University) and
  a PhD from Stanford University and was a postdoctoral fellow at Californi
 a Institute of Technology. Before joining Meta (then Facebook) in 2020\, h
 e spent 14 years as a Researcher at Microsoft Research. He currently serve
 s as an associate editor for the SIAM Journal on Optimization\, Mathematic
 al Programming\, and the Journal of Machine Learning Research. His current
  research interests include theory and algorithms for large-scale optimiza
 tion\, reinforcement learning\, and parallel and distributed computing.
GEO:40.607787;-75.381521
LOCATION:Mohler Laboratory\, 453
SUMMARY:ISE Seminar Series 
URL;VALUE=URI:https://eventscalendar.lehigh.edu/event/ise-seminar-series-25
 27
CATEGORIES:Talks & Lectures
END:VEVENT
END:VCALENDAR