RUS  ENG
Full version
JOURNALS // Informatsionnye Tekhnologii i Vychslitel'nye Sistemy // Archive

Informatsionnye Tekhnologii i Vychslitel'nye Sistemy, 2018 Issue 1, Pages 30–36 (Mi itvs291)

COMPUTING SYSTEMS

About software packages for fast automatic differentiation

A. Yu. Gorchakov

Dorodnicyn Computing Centre, FRC CSC RAS

Abstract: In the article, a comparative analysis of the packets of fast automatic differentiation Adept, CoDiPack and FADBAD is carried out. The example of the inverse coefficient problem shows the advantage of the direct method of fast automatic differentiation implemented in the CoDiPack package over the direct and inverse methods implemented in the Adept package and over the direct method implemented in the FADBAD package. As numerical optimization methods, the gradient descent method and the Levenberg-Marquardt method were used. Recommendations are given on the choice of optimization methods and rapid automatic differentiation, depending on the dimension of the problem.

Keywords: optimal control, fast automatic differentiation, gradient, Jacobi matrix, Levenberg-Marquardt method.



Bibliographic databases:


© Steklov Math. Inst. of RAS, 2024