BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Sabre//Sabre VObject 4.6.0//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:Europe/Zurich
X-LIC-LOCATION:Europe/Zurich
TZURL:http://tzurl.org/zoneinfo/Europe/Zurich
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19810329T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19961027T030000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:news235@dmi.unibas.ch
DTSTAMP;TZID=Europe/Zurich:20180716T204408
DTSTART;TZID=Europe/Zurich:20161118T110000
SUMMARY:Seminar in Numerical Analysis: Peter Zaspel (Universität Heidelber
 g / HITS)
DESCRIPTION:Hierarchical matrices approximate specific types of dense matri
 ces\,  e.g.\, from discretized integral equations\, kernel-based approxima
 tion  and Gaussian process regression\, leading to log-linear time complex
 ity  in dense matrix-vector products. To be able to solve large-scale  app
 lications\, H-matrix algorithms have to be parallelized. A special  kind o
 f parallel hardware are many-core processors\, e.g. graphics  processing u
 nits (GPUs). The parallelization of H-matrices on many-core  processors is
  difficult due to the complex nature of the underlying  algorithms that ne
 ed to be mapped to rather simple parallel operations.\\r\\nWe are interest
 ed to use these many-core processors for the full  H-matrix construction a
 nd application process. A motivation for this  interest lies in the well-k
 nown claim that future standard processors  will evolve towards many-core 
 hardware\, anyway. In order to be prepared  for this development\, we want
  to discuss many-core parallel formulations  of classical H-matrix algorit
 hms and adaptive cross approximations.\\r\\nIn the presentation\, the use 
 of H-matrices is motivated by the  model application of kernel-based appro
 ximation for the solution of  parametric PDEs\, e.g. PDEs with stochastic 
 coefficients. The main part  of the talk will be dedicated to the challeng
 es of H-matrix  parallelizations on many-core hardware with the specific m
 odel hardware  of GPUs. We propose a set of parallelization strategies whi
 ch overcome  most of these challenges. Benchmarks of our implementation ar
 e used to  explain the effect of different parallel formulations of the al
 gorithms.
X-ALT-DESC:Hierarchical matrices approximate specific types of dense matric
 es\,  e.g.\, from discretized integral equations\, kernel-based approximat
 ion  and Gaussian process regression\, leading to log-linear time complexi
 ty  in dense matrix-vector products. To be able to solve large-scale  appl
 ications\, H-matrix algorithms have to be parallelized. A special  kind of
  parallel hardware are many-core processors\, e.g. graphics  processing un
 its (GPUs). The parallelization of H-matrices on many-core  processors is 
 difficult due to the complex nature of the underlying  algorithms that nee
 d to be mapped to rather simple parallel operations.\nWe are interested to
  use these many-core processors for the full  H-matrix construction and ap
 plication process. A motivation for this  interest lies in the well-known 
 claim that future standard processors  will evolve towards many-core hardw
 are\, anyway. In order to be prepared  for this development\, we want to d
 iscuss many-core parallel formulations  of classical H-matrix algorithms a
 nd adaptive cross approximations.\nIn the presentation\, the use of H-matr
 ices is motivated by the  model application of kernel-based approximation 
 for the solution of  parametric PDEs\, e.g. PDEs with stochastic coefficie
 nts. The main part  of the talk will be dedicated to the challenges of H-m
 atrix  parallelizations on many-core hardware with the specific model hard
 ware  of GPUs. We propose a set of parallelization strategies which overco
 me  most of these challenges. Benchmarks of our implementation are used to
   explain the effect of different parallel formulations of the algorithms.
  
DTEND;TZID=Europe/Zurich:20161118T120000
END:VEVENT
END:VCALENDAR