BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20190719T085744Z
LOCATION:HG EO Nord
DTSTART;TZID=Europe/Stockholm:20190613T195000
DTEND;TZID=Europe/Stockholm:20190613T215000
UID:submissions.pasc-conference.org_PASC19_sess179_post138@linklings.com
SUMMARY:CSM13 - Scaling Iterative Random Forest for Use on Distributed Hig
 h-Performance Compute Resources
DESCRIPTION:Poster\n\n\nCSM13 - Scaling Iterative Random Forest for Use on
  Distributed High-Performance Compute Resources\n\nCliff, Romero, Brown, J
 acobson\n\nThe iterative Random Forest algorithm modifies the well-known R
 andom Forest algorithm to handle variable importance and variable-interact
 ion space in regression and classification by building sequential forests 
 of weighted decision trees. These forests are mined for recurring patterns
 . The strength of this method lies in being able to build a large number o
 f trees that provide an even larger number of decisions. While each tree c
 an be assigned to a single CPU core, allowing for the parallel computation
  of many trees, if more trees are desired than could be created at once, t
 hey must be created in batches, thus slowing down the creation of the whol
 e forest. By spreading the tree creation across distributed HPC multi-node
  systems, the forest creation can make use of large numbers of cores, mass
 ively parallelizing the creation and allowing for faster prediction and an
 alysis of decision patterns. These improvements enable research in areas t
 hat require large amounts of data at run time and contain potentially mass
 ive combinatoric effects, such as finding genetic interactions within whol
 e genomes of biofuel producers that correspond to higher yield. This work 
 shows the results of scaling using MPI and algorithmic adjustments, implem
 ented on HPC systems such as Summit and Titan.
END:VEVENT
END:VCALENDAR