BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260601T175451EDT-7974UeLejo@132.216.98.100
DTSTAMP:20260601T215451Z
DESCRIPTION:Title:\n\nHigh dimensional limit of streaming SGD for generaliz
 ed linear models\n\nAbstract: \n\nWe provide a characterization of the hig
 h dimensional limit of one-pass\, single batch stochastic gradient descent
  (SGD) in the case where the number of samples scales proportionally with 
 the problem dimension. We characterize the limiting process in terms of it
 s convergence to a high-dimensional stochastic differential equation\, ref
 erred to as the homogenized SGD. Our proofs assume Gaussian data but allow
  for a very general covariance structure. Our set-up covers a range of opt
 imization problems including linear regression\, logistic regression\, and
  some simple neural nets. For each of these models\, the convergence of SG
 D to homogenized SGD enables us to derive a close approximation of the sta
 tistical risk (with explicit and vanishing error bounds) as the solution t
 o a Volterra integral equation. In a separate paper\, we perform similar a
 nalysis without the Gaussian assumption in the case of SGD for linear regr
 ession. (Based on joint work with C. Paquette\, E. Paquette\, I. Seroussi)
 .\n
DTSTART:20230925T200000Z
DTEND:20230925T210000Z
LOCATION:Room 1104\, Burnside Hall\, CA\, QC\, Montreal\, H3A 0B9\, 805 rue
  Sherbrooke Ouest
SUMMARY:Elizabeth Collins-Woodfin (91ºÚÁÏÍø)
URL:/mathstat/channels/event/elizabeth-collins-woodfin
 -mcgill-349361
END:VEVENT
END:VCALENDAR