BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Principles and Techniques of Automatic Differentiation - Laurent H
 ascoët\, INRIA
DTSTART:20131206T100000Z
DTEND:20131206T110000Z
UID:TALK49256@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:Computing accurate derivatives of a numerical model is a cruci
 al task in many domains of Scientific Computing\, in particular for gradie
 nt-based optimization and inverse problems. Automatic Differentiation (AD)
  is a software technique to obtain derivatives of functions provided as pr
 ograms. Given a numerical model F implemented as a program P\, AD adapts o
 r transforms P into a new program that computes derivatives of F. We show 
 the mathematical formalization that both justifies AD and explains its lim
 itations. We shortly describe the software analyses that allow AD tools to
  produce more efficient code. We focus on the adjoint mode of AD\, arguabl
 y the only way to obtain gradients at a reasonable cost\, and show two rea
 l Scientific Computing applications. We give a brief panorama of current A
 D tools and conclude on research directions.
LOCATION:Auditorium\, Microsoft Research Ltd\, 21 Station Road\, Cambridge
 \, CB1 2FB
END:VEVENT
END:VCALENDAR