BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Can Federated Learning Save the Planet? - Xinchi Qiu &amp\; Nic La
 ne | University of Cambridge
DTSTART:20210427T100000Z
DTEND:20210427T113000Z
UID:TALK159049@talks.cam.ac.uk
CONTACT:87364
DESCRIPTION:Machine learning (ML) underpins a wide range of systems that w
 e all use on a daily basis. Every time we perform a web search\, use our s
 martphone or interact with a home assistant\, ML is required. However incr
 easingly\, the world is recognizing ML often has a considerable environmen
 tal cost due the energy used by powerful datacenters where it runs. For ex
 ample\, training a single state-of-the-art NLP model\, that processes text
 \, can result in CO2 emissions roughly equal to the lifetime carbon footpr
 int of five cars. More alarmingly\, for nearly a whole decade the resource
 s required by ML have been approximately doubling every three months. Alre
 ady datacenters account for 0.3% of the world's carbon emissions\; but giv
 en these existing trends -- when combined with the rapidly expanding use o
 f ML in industry\, business and personal life -- then the environmental co
 nsequences of ML must be addressed moving forward\, with a level of seriou
 sness and attention that it has never received before.\n\nTo this end\, we
  will present our work that examines if Federated Learning (FL) may offer 
 important opportunities in building future environmentally sustainable for
 ms of ML. FL is an emerging technique that enables edge devices (e.g.\, ph
 ones and embedded devices) to collaboratively learn models. The key benefi
 t of FL is traditionally grounded in user privacy\, as it allows edge devi
 ces to keeping training data on the device rather than it being shared wit
 h a datacenter controlled by a third party. But by shifting resource usage
  from the datacenter to edge devices there are also potential benefits to 
 carbon emissions\, for instance\, edge devices do not require datacenter-s
 tyle cooling (that are often 60% of total datacenter energy needs). Our ex
 ploratory research considers the type of role FL can play in future low-ca
 rbon ML solutions\, and seeks to overcome obstacles within FL itself (e.g.
 \, high communication overhead) that limit its current suitability. To the
  best of our knowledge\, this is the first time FL has been studied from t
 he perspective of its impact on environmental factors. 
LOCATION:https://zoom.us/j/6708259482?pwd=Qk03U3hxZWNJZUZpT2pVZnFtU2RRUT09
END:VEVENT
END:VCALENDAR