BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Efficient Computation through Tuned Approximation - David Keyes (K
 AUST)
DTSTART:20240516T150000Z
DTEND:20240516T160000Z
UID:TALK215842@talks.cam.ac.uk
CONTACT:Hamza Fawzi
DESCRIPTION:Numerical software is being reconstructed to provide opportuni
 ties to tune dynamically the accuracy of computation to the requirements o
 f the application\, resulting in savings of memory\, time\, and energy. Fl
 oating point computation in science and engineering has a history of “ov
 ersolving” relative to requirements or worthiness for many models. So of
 ten are real datatypes defaulted to double precision that GPUs did not gai
 n wide acceptance in simulation environments until they provided in hardwa
 re operations not required in their original domain of graphics. However\,
  driven by performance or energy incentives\, much of computational scienc
 e is now reverting to employ lower precision arithmetic where possible. Ma
 ny matrix operations considered at a blockwise level allow for lower preci
 sion and\, in addition\, many blocks can be approximated with low rank nea
 r equivalents. This leads to smaller memory footprint\, which implies high
 er residency on memory hierarchies\, leading in turn to less time and ener
 gy spent on data copying\, which may even dwarf the savings from fewer and
  cheaper flops. We provide examples from several application domains\, inc
 luding a look at campaigns in geospatial statistics and seismic processing
  that earned Gordon Bell Prize finalist status in\, resp.\, 2022 and 2023.
LOCATION:Centre for Mathematical Sciences\, MR14
END:VEVENT
END:VCALENDAR