BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Personalized collaboration for learning
DTSTART:20210826T080000
DTEND:20210826T100000
DTSTAMP:20260407T062437Z
UID:14b9711b949a98262de97faeb1f3c4594bccfa10a99f9129296d6afa
CATEGORIES:Conferences - Seminars
DESCRIPTION:El Mahdi Chayti\nEDIC candidacy exam\nexam president: Prof. Ra
 chid Guerraoui\nthesis advisor: Prof. Martin Jaggi\nco-examiner: Prof. Nic
 olas Flammarion\n\nAbstract\nIn many scenarios we have agents that collabo
 rate to learn useful representations\, such scenarios are for example Fede
 rated learning or decentralized learning in which each agent is a node and
  the goal is to train a global model which is\, in general\, a weighted av
 erage of all the available agents.  It is a very well-known fact that the
  global model does not generalize well\, in many interesting cases\, at th
 e local level\, the main reason for this is the dissimilarity between the 
 agents (agents behave differently\, biased towards their own preferences).
  This fact justified the need for personalization i.e collaborative approa
 ches that generate models that behave well at the local level. In particul
 ar\, it has been noted the similarity between this framework and the meta-
 learning / few-shot learning paradigm. Many ideas were suggested in all of
  these learning domains\, but while there is no shortage of ideas\, there 
 is clearly a need for theoretical guarantees. In this work\, we will inves
 tigate approaches to solve the personalization problem in particular and a
 nswer the more general question: how agents that do not necessarily behave
  in the same way can have a beneficial collaboration with each other?\n\nB
 ackground papers\n\nOn the Theory of Transfer Learning: The Importance of 
 Task Diversity\nhttps://arxiv.org/pdf/2006.11650.pdf\n\nWhich Tasks Should
  Be Learned Together in Multi-task Learning?\nhttps://arxiv.org/pdf/1905.0
 7553.pdf\n\nSurvey of Personalization Techniques for Federated Learning\nh
 ttps://arxiv.org/pdf/2003.08673.pdf\n\n 
LOCATION:
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR