Publication: An Enhanced Markowitz Rule for Accumulating Jacobians Efficiently
Introduction
Applications
Tools
Research Groups
Workshops
Publications
   List Publications
   Advanced Search
   Info
   Add Publications
My Account
About
Impress

An Enhanced Markowitz Rule for Accumulating Jacobians Efficiently

- Part of a collection -
 

Author(s)
Uwe Naumann

Published in
ALGORITHMY'2000 Conference on Scientific Computing

Editor(s)
K. Mikula

Year
2000

Abstract
Jacobian matrices can be accumulated using either the forward or reverse mode of Automatic Differentiation. Alternatively, derivative code can be generated to compute the Jacobian matrix directly at the current argument. The minimization of the corresponding number of arithmetic operations leads to a computationally hard combinatorial optimization problem. A new powerful heuristic for its approximate solution will be presented. The resulting codes lead to a speedup of three and more for most problems.

AD Theory and Techniques
Code Optimization

BibTeX
@INPROCEEDINGS{
         Naumann2000AEM,
       abstract = "Jacobian matrices can be accumulated using either the forward or reverse mode of
         Automatic Differentiation. Alternatively, derivative code can be generated to compute the Jacobian
         matrix directly at the current argument. The minimization of the corresponding number of arithmetic
         operations leads to a computationally hard combinatorial optimization problem. A new powerful
         heuristic for its approximate solution will be presented. The resulting codes lead to a speedup of
         three and more for most problems.",
       ad_theotech = "Code Optimization",
       author = "Uwe~Naumann",
       title = "An Enhanced {M}arkowitz Rule for Accumulating {J}acobians Efficiently",
       booktitle = "ALGORITHMY'2000 Conference on Scientific Computing",
       pages = "320--329",
       year = "2000",
       editor = "K.~Mikula",
       month = "September"
}


back
  

Contact:
autodiff.org
Username:
Password:
(lost password)