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:20260202T201804Z
LOCATION:231
DTSTART;TZID=America/Chicago:20251116T121000
DTEND;TZID=America/Chicago:20251116T123000
UID:submissions.supercomputing.org_SC25_sess223_ws_scalah108@linklings.com
SUMMARY:A High Performance GPU CountSketch Implementation and Its Applicat
 ion to Multisketching and Least Squares Problems
DESCRIPTION:Andrew Higgins, Erik Boman, and Ichitaro Yamazaki (Sandia Nati
 onal Laboratories)\n\nRandom sketching is a dimensionality reduction techn
 ique that approximately preserves norms and singular values up to some O(1
 ) distortion factor with high probability. The most popular sketches in li
 terature are the Gaussian sketch and the subsampled randomized Hadamard tr
 ansform, while the CountSketch has lower complexity. Combining two sketche
 s, known as multisketching, offers an inexpensive means of quickly reducin
 g the dimension of a matrix by combining a CountSketch and Gaussian sketch
 .\n\nHowever, there has been little investigation into high performance Co
 untSketch implementations. In this work, we develop an efficient GPU imple
 mentation of the CountSketch, and demonstrate the performance of multisket
 ching using this technique. We also demonstrate the potential for using th
 is implementation within a multisketched least squares solver that is up t
 o 77% faster than the normal equations with significantly better numerical
  stability, at the cost of an O(1) multiplicative factor introduced into t
 he relative residual norm.\n\nRecording: Livestreamed, Recorded\n\nRegistr
 ation Category: Technical Program Reg Pass, Workshop Reg Pass\n\nSession C
 hairs: Vassil Alexandrov (Hartree Centre, STFC); Jack Dongarra (University
  of Tennessee, Knoxville; Oak Ridge National Laboratory (ORNL)); Erik Drae
 ger (Lawrence Livermore National Laboratory (LLNL), Center for Applied Sci
 entific Computing); Philippa Rubin (STFC Hartree Centre); Dieter A. Kranzl
 mueller (Ludwig-Maxmilians-Universität München, Leibniz Supercomputing Cen
 tre (LRZ)); and Christian Engelmann (Oak Ridge National Laboratory (ORNL))
 \n\n
END:VEVENT
END:VCALENDAR