Christèle FAURE home page

Position
<< Chargé de recherche contractuel (until september 2000)>> in the SAFIR team until December 1999 and now in the TROPICS team at the Sophia Antipolis Research Center of INRIA.

Challenges :
Finding a position in Automatic Differentiation from September 2000.

My present interest is: studying and developing a Kernel for Automatic Differentiation to allow easy implementation of new strategies of differentiation. The present pre-compilers (source transformation) use strategies that have been designed ten years ago. Now that those strategies have been proven effective, one can think of a lot of other strategies and if a generic KAD was available implementing and testing those strategies would be easy.

Research in AD requires knowledge within: compiler technology, mathematical differentiation and usage of derivatives. This is one of the reasons why AD is so uneasy to classify within standard research classifications. Let consider that all information required to differentiate an original code are known, and that the AD users have specified the necessary derivatives, the challenge is now to generate efficient derivatives. As for the reverse mode, I have proven it effective but not optimal on large codes as shown in the report F98a. I have studied some strategies for generating adjoint codes described in F99a. Now, I try to design a KAD for implementing and evaluation those strategies on real world codes written in ocaml.

Software development
I am in charge with the development of Odyssée which is a tool-box for Automatic Differentiation developed at INRIA in the SAFIR team.
It takes as input a fortran 77 code and a list of input variables and generates a new fortran 77 set of units which computes :

• in direct mode : the tangent code (directional derivative).
• in reverse mode : the cotangent code (transposed jacobian matrix applied to a vector).

Publications

You can find my publications since 1994 within the general list of publications on Automatic Differentiation and between 1990 - 1994 at the end of this page.

Referee for ACM Transaction on Mathematical Software, and Journal of Computational Physics.

Other professional activities

• Coordination of a research action financed by INRIA called MIO (french text). People from different teams of INRIA work within this action. MIO stands for operational reverse mode and aims at making the reverse mode of Automatic Differentiation practicable on operational codes.

• Coordination of the Automatique Differentiation activity in European Projects : DECISION (HPCN)
  Subject : multi-criteria/multidisciplinary optimization
  Partnership : DASSAULT Aviation, MESSET, VTT, NOKKA-TUME, Nag, INRIA, JYVASKYLA University.

• Realisation of the prototype of a Web site that presents briefly INRIA research activities for non specialists. (These Web pages are not yet approuved by INRIA and must be considered as a personnal presentation.)

Organization
Coorganization of the third workshop on Automatic Differentiation AD2000 in Nice in the first semester of 2000.