BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Strategyproof and Byzantine Resilient Personalized Collaborative L
 earning
DTSTART:20210713T140000
DTEND:20210713T160000
DTSTAMP:20260407T100420Z
UID:e5a7bcbb139270862b2944dce476caf24dd5296c213087d0516651a3
CATEGORIES:Conferences - Seminars
DESCRIPTION:Sadegh Farhadkhani\nEDIC candidacy exam\nexam president: Prof.
  Emmanuel Abbé\nthesis advisor: Prof. Rachid Guerraoui\nco-examiner: Prof
 . Nicolas Flammarion\n\nAbstract\nCollaborative learning is a proposed fra
 mework that enables different nodes to learn from each other's data. In mo
 dern applications\, this is very appealing\, as the dataset of a single no
 de is often too small to achieve acceptable performances. However\, in thi
 s setting\, Byzantine or strategic nodes may deviate from the protocol in 
 order to sabotage the learning procedure or bias the models learned by oth
 er nodes in their favor. The latter issue can particularly be the case in 
 recommender systems\, where there are significant social and economic ince
 ntives to promote certain viewpoints and products. Therefore\, we aim to d
 esign a personalized collaborative learning framework that is both Byzanti
 ne resilient and strategyproof.\nIn this report\, first\, we introduce a n
 ew formulation for federated learning that enables each node to learn a pe
 rsonalized model while using other nodes' data. Second\, we discuss a Byza
 ntine resilient distributed learning algorithm that is robust to malicious
  workers sending forged gradient estimates. Third\, we introduce a linear 
 regression framework that incentivizes honesty for the strategic nodes tha
 t want to pull the regression model towards their own desired points. Fina
 lly\, we discuss our research plans.\n\nBackground papers\n\n	Machine Lear
 ning with Adversaries: Byzantine Tolerant Gradient Descent [Link]\n	Federa
 ted Learning of a Mixture of Global and Local Models [Link]\n	Strategyproo
 f Linear Regression in High Dimensions [Link]\n
LOCATION:
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR