BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260602T064140EDT-8854vNaH1D@132.216.98.100
DTSTAMP:20260602T104140Z
DESCRIPTION:Title:\n\nConstrained and Multirate Training of Neural Networks
 \n\nAbstract: \n\nI will describe algorithms for regularizing and training
  deep neural networks. Soft constraints\, which add a penalty term to the 
 loss\, are typically used as a form of explicit regularization for neural 
 network training. In this talk I describe a method for efficiently incorpo
 rating constraints into a stochastic gradient Langevin framework for the t
 raining of deep neural networks. In contrast to soft constraints\, our con
 straints offer direct control of the parameter space\, which allows us to 
 study their effect on generalization. In the second part of the talk\, I w
 ill focus on the role played by individual layers and substructures of neu
 ral networks. In particular\, I will show that by choosing appropriate par
 titionings of the network parameters into fast and slow parts\, a multirat
 e approach can be used to train deep neural networks for transfer learning
  applications in vision and natural language processing in half the time\,
  without reducing the generalization performance of the model.\n
DTSTART:20230911T200000Z
DTEND:20230911T210000Z
LOCATION:Room 1104
SUMMARY:Medhi Dagdoug (91ºÚÁÏÍø Univesity)
URL:/mathstat/channels/event/medhi-dagdoug-mcgill-univ
 esity-350550
END:VEVENT
END:VCALENDAR