Programme of the Eighth Euro AD Workshop

Friday, July 17, 2009

• 8³⁰ –9⁰⁰   Registration
• 9⁰⁰ –9¹⁵   Welcome
• 9¹⁵ –9⁴⁵   Talk
  1. Hans Skaug (University of Bergen)
     The AD Model Builder project
     + Abstract
     
     AD Model Builder (ADMB) is a powerful programming package that uses automatic differention for the development and fitting of statistical models to data. It contains a suite of techniques (e.g. powerful function minimizer, profile likelihoods, MCMC simulation, random effects modelling) which makes it a unique framework for working with highly parameterized nonlinear models. ADMB was originally developed by David Fournier and is now an open source project (http://admb-project.org/). It has found application in numerous practical natural resource assessment and management projects has been used in hundreds of scientific papers. In this talk we will briefly introduce the internal working of ADMB as the development of user applications. We also describe the open source project, the goal of which is to maintain and further develop AD Model Builder.
• 9⁴⁵ –10¹⁵   Talk
  2. Finlay Scott (CEFAS)
     AD in R
     + Abstract
     
     R is a powerful, flexible and open source software environment for statistical computing and graphics. It is widely used throughout the scientific community and its popularity is growing. One of R's key advantages is that it is highly extensible. There are currently about 2000 user-developed packages publicly available that expand the core functionality. It is also possible to link R to other languages, including C/C++ and Fortran, to speed up computationally intensive processes. Although R includes general purpose optimisation routines based on numerical approximation and also some basic symbolic differentiation, R does not currently include Automatic Differentiation (AD) capabilities. We have linked R to open-source AD libraries such as ADOL-C and Autodiff to develop fisheries models, demonstrating the value of using AD within R. However, this approach has focused on specific applications and there is currently no generic AD interface for R. Here we present a general overview of the R language and highlight the importance of developing such a generic, useable AD interface for R. We are interested in collaborating with the AD community to help us achieve this goal.
• 10¹⁵ –10⁴⁵   Talk
  3. Sebastian Walter (HU Berlin)
     AD in Python with Application in Science and Engineering
     + Abstract
     
     Python becomes more and more popular as a general purpose programming language, also in science and engineering. In this talk we show a few examples of scientific programs that require derivative information. One example will be the optimization of an NLP with PYIPOPT. We use the examples to illustrate how to differentiate those programs with PyCPPAD, PyADOLC and ALGOPY from the user's perspective. We also discuss the execution speed of PyADOLC and ALGOPY with pure C++ ADOL-C and briefly explain how the interface between Python and C++ is implemented with Boost::Python.
• 10⁴⁵ –11¹⁵   Coffee
• 11¹⁵ –11⁴⁵   Talk
  4. Laurent Hascoet (INRIA Sophia-Antipolis, France)
     Tapenade as a front-end for ADOL-C
     + Abstract
     
     Our team is designing a special mode of Tapenade that makes it a preprocessor for ADOL-C. After Tapenade data-flow and activity analysis, we generate source for ADOL-C with selected variable types changed to "aDouble". We shall discuss some issues raised by this development.
• 11⁴⁵ –12¹⁵   Talk
  5. Jean Utke (Argonne National Laboratory)
     Rapsodia
• 12¹⁵ –13⁴⁵   Lunch
• 13⁴⁵ –14¹⁵   Talk
  6. Paul Hovland (Argonne National Laboratory)
     TBD
     + Abstract
     
     TBD
• 14¹⁵ –14⁴⁵   Talk
  7. Viktor Mosenkis (RWTH Aachen)
     On integer stack compression in adjoint codes
• 14⁴⁵ –15¹⁵   Coffee
• 15¹⁵ –16⁰⁰   Talk
  8. Jeremy Walton (NAG Ltd.)
     NAG and Automatic Differentiation
• 16⁰⁰ –17⁰⁰   Open Discussion and Closing Remarks
• 17⁰⁰   The End