BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Hey\, you got your distributed algorithm in my ML! - Tim Harris\, 
 Microsoft
DTSTART:20210715T140000Z
DTEND:20210715T150000Z
UID:TALK160909@talks.cam.ac.uk
CONTACT:Srinivasan Keshav
DESCRIPTION:I started work in systems for machine learning around a year a
 go\, joining the AI Frameworks team at Microsoft\, and working from infere
 nce on small client devices up to model training over distributed GPU clus
 ters.  \n\nI am going to talk about some of the things that surprised me. 
  Central to this is the difficulty in identifying abstractions.  We have t
 echniques such as pipeline parallelism and tensor parallelism\, but the im
 plementations are tightly coupled with the models themselves\, and optimiz
 ed alongside them.  It is tempting to design fresh alternatives that can d
 ecouple distribution techniques from models\; however\, deploying these in
 to an evolving ecosystem is itself a challenge.\n\nI will wrap up by descr
 ibing some of our recent work to chart an incremental path through these t
 opics.\n\nShort Bio: I am a Principal Architect at Microsoft\, focused on 
 PyTorch and the ONNX runtime. Prior to that I was with AWS and worked on l
 arge-scale storage performance and data analytics with Amazon S3. Further 
 back\, I led the Oracle Labs group in Cambridge\, UK working on runtime sy
 stems for in-memory graph analytics\, and the confluence of work on “big
  data” and ideas from high-performance computing. Before joining Oracle 
 I was with Microsoft in a prior stint (2004–2012)\, and on the faculty o
 f the University of Cambridge Computer Laboratory (2000–2004) during the
  early days of the Xen hypervisor project.
LOCATION:https://cl-cam-ac-uk.zoom.us/j/4361570789?pwd=Nkl2T3ZLaTZwRm05bzR
 TOUUxY3Q4QT09&from=addon
END:VEVENT
END:VCALENDAR