skip to main content
DOE PAGES title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Cheaper Adjoints by Reversing Address Computations

Abstract

The reverse mode of automatic differentiation is widely used in science and engineering. A severe bottleneck for the performance of the reverse mode, however, is the necessity to recover certain intermediate values of the program in reverse order. Among these values are computed addresses, which traditionally are recovered through forward recomputation and storage in memory. We propose an alternative approach for recovery that uses inverse computation based on dependency information. Address storage constitutes a significant portion of the overall storage requirements. An example illustrates substantial gains that the proposed approach yields, and we show use cases in practical applications.

Authors:
 [1];  [2];  [3]
  1. INRIA Sophia-Antipolis, 2004 Route des lucioles, BP 93 06902 Valbonne, France
  2. Argonne National Laboratory, MCS, Argonne, IL 60439, USA
  3. Department of Computer Science, RWTH Aachen University, 52056 Aachen, Germany
Publication Date:
Sponsoring Org.:
USDOE
OSTI Identifier:
1197992
Grant/Contract Number:  
AC02-06CH11357
Resource Type:
Published Article
Journal Name:
Scientific Programming
Additional Journal Information:
Journal Name: Scientific Programming Journal Volume: 16 Journal Issue: 1; Journal ID: ISSN 1058-9244
Publisher:
Hindawi Publishing Corporation
Country of Publication:
Egypt
Language:
English

Citation Formats

Hascoët, L., Utke, J., and Naumann, U. Cheaper Adjoints by Reversing Address Computations. Egypt: N. p., 2008. Web. doi:10.1155/2008/375243.
Hascoët, L., Utke, J., & Naumann, U. Cheaper Adjoints by Reversing Address Computations. Egypt. doi:10.1155/2008/375243.
Hascoët, L., Utke, J., and Naumann, U. Tue . "Cheaper Adjoints by Reversing Address Computations". Egypt. doi:10.1155/2008/375243.
@article{osti_1197992,
title = {Cheaper Adjoints by Reversing Address Computations},
author = {Hascoët, L. and Utke, J. and Naumann, U.},
abstractNote = {The reverse mode of automatic differentiation is widely used in science and engineering. A severe bottleneck for the performance of the reverse mode, however, is the necessity to recover certain intermediate values of the program in reverse order. Among these values are computed addresses, which traditionally are recovered through forward recomputation and storage in memory. We propose an alternative approach for recovery that uses inverse computation based on dependency information. Address storage constitutes a significant portion of the overall storage requirements. An example illustrates substantial gains that the proposed approach yields, and we show use cases in practical applications.},
doi = {10.1155/2008/375243},
journal = {Scientific Programming},
number = 1,
volume = 16,
place = {Egypt},
year = {2008},
month = {1}
}

Journal Article:
Free Publicly Available Full Text
Publisher's Version of Record
DOI: 10.1155/2008/375243

Save / Share: