BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:AI Center Seminar - AI Fundamentals series - Dr. Denny Wu
DTSTART:20250625T151500
DTEND:20250625T161500
DTSTAMP:20260316T095539Z
UID:16a2d0491c36152a7404a1bdb329e269b25a32272f4bf72d05985fcf
CATEGORIES:Conferences - Seminars
DESCRIPTION:Denny Wu\nThe talk is jointly organized by the EPFL Foundatio
 ns of Learning and AI Research (FLAIR) group and the EPFL AI Center.\n\
 nHosting professor: Prof. Lenka Zdeborova\n\nFor logistics purposes\, plea
 se register here (using your EPFL email address): HERE. \nThe talk will b
 e followed by a coffee session. \n\nTitle\nLearning shallow neural networ
 ks in high dimensions: SGD dynamics and scaling laws\n\nAbstract\nWe study
  the sample and time complexity of online stochastic gradient descent (SGD
 ) in learning a two-layer neural network with M orthogonal neurons on isot
 ropic Gaussian data. We focus on the challenging “extensive-width” reg
 ime M>>1 and allow for large condition number in the second-layer paramete
 rs\, covering the power-law scaling a_m= m^{-\\beta} as a special case. W
 e characterize the SGD dynamics for the training of a student two-layer ne
 ural network and identify sharp transition times for the recovery of each 
 signal direction. In the power-law setting\, our analysis entails that whi
 le the learning of individual teacher neurons exhibits abrupt phase transi
 tions\, the juxtaposition of emergent learning curves at different timesca
 les results in a smooth scaling law in the cumulative objective. \n\nBio\
 nDenny Wu is a Faculty Fellow at the Center for Data Science\, New York Un
 iversity and the Center for Computational Mathematics\, Flatiron Institu
 te. He was a Ph.D. student at the University of Toronto and the Vector Ins
 titute\, advised by Jimmy Ba and Murat A. Erdogdu. Prior to that\, he comp
 leted his undergraduate studies at Carnegie Mellon University supervised b
 y Ruslan Salakhutdinov. \n 
LOCATION:ELE 117 https://plan.epfl.ch/?room==ELE%20117 https://epfl.zoom.u
 s/j/69657848717?pwd=Rx2qTxGyWsYDWJg7wERlVkczgB0nBq.1
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR