Friday, May 14

Automatic Differentiation and Parallel Processing

10:30 AM-12:30 PM
Room: Atlanta 5

The popularity of automatic differentiation (AD) as a tool for scientific computing has increased dramatically over the past decade. During that same period, the power and availability of parallel machines have experienced tremendous growth. A natural consequence has been research on the interplay between AD and parallel computations. This research has addressed not only how to apply AD to explicitly parallel programs but how to exploit the new potential for parallelism introduced by differentiation. The speakers will provide an introduction to the technique of AD, summarize the results of the research on AD and parallelism, and describe opportunities for continued research.

Organizer: Paul D. Hovland
Argonne National Laboratory

10:30-10:55 Overview
Paul D Hovland, Organizer
11:00-11:25 ADIFOR-MP: Automatic Differentiation of Message Passing Parallel Programs
Alan Carle, Rice University
11:30-11:55 Exploiting AD-enabled Time Parallelism in the Reverse Mode
Andreas Griewank, Technical University of Dresdren, Germany
12:00-12:25 Exploiting AD-enabled Parallelism via Multithreading
Lucas Roh, Argonne National Laboratory

AN99 Home


Program Updates

Speaker Index




LMH, 1/20/99, MMD, 1/27/99