BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Disciplined Approximate Computing: From Language to Hardware and B
 eyond - Luis Ceze - University of Washington.
DTSTART:20131023T131500Z
DTEND:20131023T141500Z
UID:TALK47033@talks.cam.ac.uk
CONTACT:David Greaves
DESCRIPTION:Energy is increasingly a first-order concern in computer syste
 ms. Exploiting\nenergy-accuracy trade-offs is an attractive choice in appl
 ications that can\ntolerate inaccuracies.  A key challenge\, though\, is h
 ow to isolate parts of the\nprogram that must be precise from those that c
 an be approximated so that a\nprogram functions correctly even as quality 
 of service degrades. Addressing\nthat challenge leads to opportunities for
  approximate computing across the\nentire system stack.\n\nIn this talk I 
 will describe our effort on co-designing language\,\nhardware and system s
 upport\nto take advantage of approximate computing across the system stack
  in\na safe and efficient way.\nWe use type qualifiers to declare data tha
 t may be subject to approximate\ncomputation. Using these types\, the syst
 em automatically maps approximate\nvariables to potentially imprecise and 
 unreliable but much more\nefficient storage and\ndata operations\, as well
  as more energy-efficient algorithms provided\nby the programmer.\nIn addi
 tion\, the system can statically guarantee isolation of the\nprecise progr
 am component\nfrom the approximate component. This allows a programmer to 
 control explicitly\nhow information flows from approximate data to precise
  data. Importantly\,\nemploying static analysis eliminates the need for dy
 namic checks\, further\nimproving energy savings. I will describe a micro-
 architecture that offers\nexplicit approximate storage and computation and
  a proposal on using\nneural networks as approximate accelerators for gene
 ral programs. I will\nconclude with an overview of our current/future rese
 arch directions\, including\nlanguage extensions for quality-of-result spe
 cification\, programming tools\,\napproximate persistent storage and appro
 ximate wireless communication.
LOCATION:Lecture Theatre 1\, Computer Laboratory
END:VEVENT
END:VCALENDAR