BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:AI Hardware and Real-World AI - Andrew Fitzgibbon - Distinguished 
 Engineer\, Graphcore
DTSTART:20220525T140500Z
DTEND:20220525T145500Z
UID:TALK173720@talks.cam.ac.uk
CONTACT:Ben Karniely
DESCRIPTION:AI is fast becoming a significant consumer of the world’s co
 mputational power\, so it is crucial to use that power wisely and efficien
 tly.  Our approaches to doing so must span all levels of the research stac
 k: from fundamental theoretical understanding of the loss surfaces and reg
 ularization properties of machine learning models\, to efficient layout at
  the transistor level of floating-point multipliers and RAM.  I will talk 
 about projects\, such as real-time computer vision on the Microsoft HoloLe
 ns HPU (about 3.5 GFLOPS)\, which required extreme efficiency in both obje
 ctive and gradient computations\, and how this relates to the training of 
 massive AI models on Graphcore’s IPU (about 350 TFLOPS).  Key to this wo
 rk is how we empower programmers to communicate effectively with such hard
 ware\, and how we design frameworks and languages to ensure we can put the
 ory into practice.  \n\nSo this talk contains aspects of: mathematical opt
 imization\, automatic differentiation\, programming languages\, and silico
 n design.   Despite this range of topics\, the plan is for it to be access
 ible and useful to anyone who loves computers.\n\nLink to join virtually: 
 https://cl-cam-ac-uk.zoom.us/j/97767639783?pwd=T09GcVJxZUNEUFEvRnZnbWwxeEw
 zQT09\n\nA recording of this talk is available at the following link: http
 s://www.cl.cam.ac.uk/seminars/wednesday/video/
LOCATION:Lecture Theatre 1\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR