Automatic Differentiation for Iterative Process and Its Applications in Network Performance Analysis
by
Mingyan Liu
University of Maryland, Department of Electrical and Computer Engineering
Coauthors: John S. Baras (University of Maryland)

In this paper we focus on the application of automatic differentiation (AD) technique on a network performance analysis model defined as an iterative process. When using AD, what was actually computed may differ significantly from the derivatives one intended to compute. This is especially true in the iterative evaluation of a function defined implicitly or otherwise. On the other hand, many engineering problems are impossible or impractical to be modeled explicitly, and we have to turn to approximations which often take an iterative form. We review some of the results on the convergence of general iterative processes and the convergence of the derivative code of such iterative processes. We are especially interested in a class of fixed point iteration problems and we extend some of these results to discuss this class of problems. Finally we apply the AD package ADIC to a network performance evaluation problem for numerical experiments to get sensitivities of network blocking probabilities with respect to network offered traffic load.

Welcome to Mingyan Liu's Homepage

Date received: February 11, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cads-69.