BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:2.0
BEGIN:VEVENT
DTSTART:20151119T170000Z
DTEND:20151119T173000Z
LOCATION:18CD
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Low-rank matrix approximations play an important role=0Ain a wide range of applications. To compute a low-rank=0Aapproximation of a dense matrix, a common approach uses=0Athe QR factorization with column pivoting (QRCP).=0AWhile reliable and efficient, this deterministic approach requires=0Acostly communication, which is becoming=0Aincreasingly expensive on modern computers. We use=0Aan alternative approach based on random sampling, =0Awhich requires much less communication than QRCP.=0AIn this paper, we compare the performance=0Aof the random sampling with that of QRCP on the=0ANVIDIA Kepler GPU. Our performance results demonstrate=0Athat the random sampling method can be up to 13 times faster=0Acompared to QRCP while computing=0Aan approximation of comparable accuracy. We also present the=0Aparallel scaling of random sampling over multiple GPUs, =0Ashowing a speedup of 5.1 over three GPUs.=0AThese results demonstrate the potential of the random sampling as=0Aan excellent computational tool for many applications.
SUMMARY:Performance of Random Sampling for Computing Low-Rank Approximations of a Dense Matrix on GPUs
PRIORITY:3
END:VEVENT
END:VCALENDAR