BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Federated Learning - Siddharth Swaroop (University of Cambridge)
DTSTART:20200311T110000Z
DTEND:20200311T123000Z
UID:TALK141058@talks.cam.ac.uk
CONTACT:75379
DESCRIPTION:The goal of federated learning is to perform distributed train
 ing without centralising data. The training dataset is split across multip
 le devices or clients\, potentially in a non-IID way. The key motivating f
 actors are security and privacy of personal data. Use-cases include traini
 ng on data from customers' mobile phones\, and collaboratively training a 
 healthcare model on patient data from hospitals. Communication is usually 
 the limiting factor\, and the field has grown recently with many communica
 tion-efficient algorithms proposed. We will start by considering federated
  learning's cryptographic origins. We will then explicitly consider its co
 re challenges. We will build up from the simplest proposed SGD algorithms 
 to more recent Bayesian algorithms. We will also briefly consider techniqu
 es that enhance security (like Secure Multi-Party Computation) and privacy
  (like Differential Privacy)\, which are of crucial importance in federate
 d learning. 
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR