BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260522T150117Z
LOCATION:D171
DTSTART;TZID=America/Chicago:20181114T103000
DTEND;TZID=America/Chicago:20181114T110000
UID:submissions.supercomputing.org_SC18_sess271_exforum105@linklings.com
SUMMARY:Productive and Performant AI Platforms of the Future
DESCRIPTION:Rangan Sukumar (Cray Inc)\n\nThus far, contributions to hardwa
 re and software tools for advanced analytics and artificial intelligence h
 as come from the commodity/cloud computing community. In this talk, we sha
 re exciting results from efforts that ported software frameworks such as A
 pache Spark and TensorFlow onto high performance computing (HPC) hardware 
 to make the case that HPC-approaches future-proof AI investments. We will 
 demonstrate performance gains from combining a HPC interconnect with algor
 ithmic cleverness using communication collectives for graph analytics, dee
 p learning and matrix methods – all components of the modern data science 
 and enterprise AI workflows. Based on experience from several use-cases, w
 e will argue how HPC futureproofs investments for AI journey – particularl
 y around emerging requirements around (I) non-traditional data (graphs, me
 dical imagery, genomic sequences); (ii) building custom domain-specific mo
 dels with hyper-parameter learning; (iii) the need for ensemble and model 
 parallelism (iv) latency on edge-devices and training cadence with custom 
 processors.\n\nTag: Machine Learning, Scientific Computing\n\nSession Chai
 r: Scott Michael (Indiana University)\n\n
END:VEVENT
END:VCALENDAR