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:20230831T095746Z
LOCATION:Schwarzhorn
DTSTART;TZID=Europe/Stockholm:20230627T143000
DTEND;TZID=Europe/Stockholm:20230627T150000
UID:submissions.pasc-conference.org_PASC23_sess183_pap125@linklings.com
SUMMARY:Mixed-Precision Random Projection for RandNLA on Tensor Cores
DESCRIPTION:Paper\n\nHiroyuki Ootomo and Rio Yokota (Tokyo Institute of Te
 chnology)\n\nRandom projection can reduce the dimension of data while capt
 uring its structure and is a fundamental tool for machine learning, signal
  processing, and information retrieval, which deal with a large amount of 
 data today. RandNLA (Randomized Numerical Linear Algebra) leverages random
  projection to reduce the computational complexity of low-rank decompositi
 on of tensors and solve least-square problems. While the computation of th
 e random projection is a simple matrix multiplication, its asymptotic comp
 utational complexity is typically larger than other operations in a RandNL
 A algorithm. Therefore, various studies propose methods for reducing its c
 omputational complexity. We propose a fast mixed-precision random projecti
 on method on NVIDIA GPUs using Tensor Cores for single-precision tensors. 
 We exploit the fact that the random matrix requires less precision, and de
 velop a highly optimized matrix multiplication between FP32 and FP16 matri
 ces -- SHGEMM (Single and Half GEMM) -- on Tensor Cores, where the random 
 matrix is stored in FP16. Our method can compute Randomized SVD 1.28 times
  faster and Random projection high order SVD 1.75 times faster than baseli
 ne single-precision implementations while maintaining accuracy.\n\nDomain:
  Engineering, Computer Science, Machine Learning, and Applied Mathematics 
 
\n\nSession Chair: Johannes Gebert (High-Performance Computing Center Stu
 ttgart, University of Stuttgart)
END:VEVENT
END:VCALENDAR