BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Efficient Data Movement for Machine Learning
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260609T154956Z
UID:tag:localist.com\,2008:EventInstance_47814327559624
DTSTART:20241024T183000Z
DTEND:20241024T193000Z
DESCRIPTION:Title: Efficient Data Movement for Machine Learning\n\n\nAbstra
 ct: Over the past decade\, advances in machine learning\nalgorithms and mo
 dels have enabled some remarkable applications.\nHowever\, these applicati
 ons place considerable demands on our\ncomputing infrastructure\, incurrin
 g significant equipment costs and\nprocessing delays. ML These development
 s have renewed the focus on\noptimizing the underlying systems that dictat
 e how efficiently models\ncan be trained and deployed. A vital aspect of s
 ystem efficiency is\nthe efficient movement of data. ML workloads are inte
 nsely\ndata-driven\, requiring vast amounts of data to be fed to accelerat
 ors\n(GPUs\, TPUs\, etc.) for processing. Bottlenecks in data transfer\,\n
 whether between multiple accelerators or between memory and the\naccelerat
 or severely limit performance. Minimizing latency and\nmaximizing bandwidt
 h through optimized interconnects and efficient\nscheduling of compute and
  communications are crucial for improving ML\nefficiency. In this talk\, I
  will describe our recent work on efficient\norchestration of data movemen
 t in training and inference settings and\ndemonstrate how new ways of util
 izing the hardware infrastructure can\nyield significant performance gains
 .\n\n\nSpeaker bio: Arvind Krishnamurthy is the Short-Dooley Professor in 
 the Paul G.\nAllen School of Computer Science & Engineering. His research 
 interests\nare in building effective and robust computer systems in the co
 ntext\nof both data centers and Internet-scale systems. More recently\, hi
 s\nresearch has focussed on programmable networks and systems for machine\
 nlearning. He is an ACM fellow\, a past program chair of ACM SIGCOMM and\n
 Usenix NSDI\, is a former Vice President of Usenix\, and has served on\nth
 e CRA board.
LOCATION:Science and Engineering Complex (SEC)\, LL2.224
SUMMARY:Efficient Data Movement for Machine Learning
URL;VALUE=URI:https://calendar.college.harvard.edu/event/efficient-data-mov
 ement-for-machine-learning
END:VEVENT
END:VCALENDAR