## On automatic differentiation of computer codes

Computing Science Dept.

University of Aberdeen

Monday, 3 September 2007, 10:00 (note the unusual weekday and time!)

Cybernetica Bldg (Akadeemia tee 21), room B101

Slides from the talk [pdf]

**Abstract**: In scientific computation, there is a frequent
need to compute first derivatives of a function represented by a
computer program. One way to achieve this is to use Automatic
Differentiation (AD) which allows for the computation of derivatives
of a function represented by a computer program. This is carried out
by a semantics augmentation process transforming the input code to a
new one that computes the function as well as its derivative.

In this talk, I will introduce the AD technique and I will motivate
it through an application from computational engineering. Then, I will
present an overview of tools from sparse matrix technology and
compiler optimisation techniques aimed at enhancing the performance of
the transformed code. I will also discuss the evaluation of these
techniques in the AD context.

Tarmo Uustalu

Last update 3.9.2007