From: alex@math-info.univ-paris5.fr
Date: Sat, 4 Mar 95 16:44:40 +0100
------------------------------------------------------------------------------
Name: Alex Aussem

Organization: 
European Southern Observatory, Munich, Germany &&
Universite Rene Descartes, Paris, France.

Interests: Time series prediction, recurrent neural networks, learning
procedures, dynamical system modeling and identification. Prediction
of atmospheric parameters relevant for astronomical observations with
Marc Sarazin (MET-AI list).

Projects/systems: 

1. Automatic Astronomical Weather Station for the ESO-Very Large
Telescope
People involved: Marc Sarazin, Fionn Murtagh (Statistical Prediction)
                 Alex Aussem (Neural Nets)

Description: 
The VLT-AWS (Astronomical Weather Station) is an automated system
composed of various sensors (local meteorology, optical seeing,
scintillation, precipitable water vapour, extinction) associated
to a set of predictive algorithms to depict and forecast the
observing environment of an astronomical observatory. The forecast
is used by the dynamical scheduler in charge of optimizing the
observing sequence.

Dynamical and recurrent neural networks are used to capture the
dynamic of the underlying process generating the series. Properties
inherent to the series (Lyapunov exponent, embedding dimension, etc.)
can be derived by observing the long term iterated predictions made by
the network operating in closed-loop fashion.

Publications:

A. Aussem 1992, ``Application of genetic algorithms as an adaptive model
for water demands,'' internal report (in French), Institut National
des T\'el\'ecommunications, Evry, France.

A. Aussem, F. Murtagh and M. Sarazin 1995, ``Dynamical recurrent
neural networks and pattern recognition methods for time series
prediction: application to seeing and temperature forecasting in the
context of ESO's VLT Astronomical Weather Station," {\em Vistas in
Astronomy} {\bf 38}, pp. 357-274.

A. Aussem, F. Murtagh and M. Sarazin 1995, ``Dynamical recurrent
neural networks towards environmental time series prediction,'' {\em
International Journal on Neural Systems}, accepted for publication.

F. Murtagh, A. Aussem and M. Sarazin 1995, ``Nowcasting astronomical
seeing: towards an operational approach," {\em Publications of the
Astronomical Society of the Pacific}, accepted for publication.

A. Aussem, F. Murtagh and M. Sarazin 1995, ``Fuzzy astronomical seeing
nowcasts with a dynamical and recurrent connectionist network," {\em
Neurocomputing}, submitted.

A. Aussem 1995, ``Training dynamical recurrent neural networks with
the temporal recurrent back-propagation algorithm: application to time
series prediction and characterization," IEEE {\em Transactions on
Neural Networks}, submitted.