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:20181221T160728Z
LOCATION:D161
DTSTART;TZID=America/Chicago:20181112T140000
DTEND;TZID=America/Chicago:20181112T142000
UID:submissions.supercomputing.org_SC18_sess158_ws_lasalss104@linklings.co
 m
SUMMARY:Iterative Randomized Algorithms for Low Rank Approximation of Tera
 scale Matrices with Small Spectral Gaps
DESCRIPTION:Workshop\nAlgorithms, Heterogeneous Systems, Resiliency, Works
 hop Reg Pass\n\nIterative Randomized Algorithms for Low Rank Approximation
  of Terascale Matrices with Small Spectral Gaps\n\nIyer, Gittens, Carother
 s, Drineas\n\nRandomized approaches for low rank matrix approximations hav
 e become popular in recent years and often offer significant advantages ov
 er classical algorithms because of their scalability and numerical robustn
 ess on distributed memory platforms. We present a distributed implementati
 on of randomized block iterative methods to compute low rank matrix approx
 imations for dense tera-scale matrices. We are particularly interested in 
 the behavior of randomized block iterative methods on matrices with small 
 spectral gaps. Our distributed implementation is based on four iterative a
 lgorithms: block subspace iteration, the block Lanczos method, the block L
 anczos method with explicit restarts, and the thick-restarted block Lanczo
 s method. We analyze the scalability and numerical stability of the four b
 lock iterative methods and demonstrate the performance of these methods fo
 r various choices of the spectral gap. Performance studies demonstrate sup
 erior runtimes of the block Lanczos algorithms over the subspace power ite
 ration approach on (up to) 16,384 cores of AMOS, Rensselaer's IBM Blue Gen
 e/Q supercomputer.
URL:https://sc18.supercomputing.org/presentation/?id=ws_lasalss104&sess=se
 ss158
END:VEVENT
END:VCALENDAR