SAVE - Simulated Annealing Applied to the Vertex Elimination Problem in Computational Graphs

Uwe Naumann 1
1 TROPICS - Program transformations for scientific computing
CRISAM - Inria Sophia Antipolis - Méditerranée
Abstract : The chain rule - fundamental for Automatic Differentiation (AD) - can be applied to computational graphs representing vector functions in arbitrary orders resulting in different operations counts for the calculation of their Jacobian matrices. Very few authors have looked at this interesting subject so far and there is no generally accepted terminology for dealing with these combinations of the forward and reverse modes of AD. The minimizati- on of the number of arithmetic operations required for the calculation of the complete Jacobian leads to a computationally hard combinatorial optimization problem. In this paper we will describe a new heuristic approach to the solution of the vertex elimination problem in computational graphs which can easily adapted to the more general case of eliminating edges. Simulated annealing is widely regarded as a suitable method for solving combinatorial optimization problems. We will discuss different annealing schedules and present some test results. Finally we will regard this approach in comparison with other methods for computing Jacobians using a minimal number of arithmetic operations.
Type de document :
Rapport
RR-3660, INRIA. 1999
Liste complète des métadonnées

https://hal.inria.fr/inria-00073012
Contributeur : Rapport de Recherche Inria <>
Soumis le : mercredi 24 mai 2006 - 11:35:20
Dernière modification le : samedi 27 janvier 2018 - 01:31:04
Document(s) archivé(s) le : dimanche 4 avril 2010 - 21:02:56

Fichiers

Identifiants

  • HAL Id : inria-00073012, version 1

Collections

Citation

Uwe Naumann. SAVE - Simulated Annealing Applied to the Vertex Elimination Problem in Computational Graphs. RR-3660, INRIA. 1999. 〈inria-00073012〉

Partager

Métriques

Consultations de la notice

109

Téléchargements de fichiers

135