Course in “Identification and Forecasting of Dynamical Systems by Neural Networks - Principles Techniques, Applications”

in Vasteras

Septemeber 15-17

(in connection with SIMS 2003)

 

Please send your registration before Sept 8

 

Monday 15 Sep 9 – 16 Malardalen Univ, Vasteras room S3-902

Tuesday 16 Sep  9 – 16 Malardalen Univ, Vasteras room R1-218

Wednesday 17 Sep 9 – 15 Malardalen Univ, Vasteras room S3-908

 

Lecturer:

 

                 Dr. Hans Georg Zimmermann

             Senior Principal Research Scientist

          Siemens AG, Corporate Technology, Munich

 

Course content:

 

 0 Introduction to Neural Networks

 1 Neural Algorithms: More than the Numerics of Gradient Computation

 2 Feedforward Neural Networks: More than Function Approximation

 3 Model Building: More than Learning from Data

 4 Neuro - Fuzzy: More than Neuro & Fuzzy

 5 Recurrent Neural Networks: More than Algorithms

 6 Open Systems: More than a Superposition of Internal & External

Dynamics

 7 Error Correction Neural Networks: More than Autoregressive Modeling

 8 Variance-Invariance Separation: More than Dimensionality Reduction

 9 Unfolding in Space and Time: More than Unfolding in Time

10 Time in Time Series Analysis: More than Data Time

11 Stochastic Modeling: More than Deterministic Forecasting

12 Causal-Retro-Causal Networks: More than Causal Networks

13 Online Learning: More than Plasticity versus Stability

14 Large Networks: More than Increasing Dimensionality

15 Decision Support Systems: More than Forecasting

16 Multi-Agent Market Modeling: More than Econometrics

 

In total, the lecture contains about 300 slides.

 

The part about feedforward networks is at least partly covered in the

book chapter 'How to Train Neural Networks'.

 

The part about recurrent networks is partly covered in the book chapter

'Modeling of Dynamical Systems by Error Correction Neural Networks'.

 

For Portfolio Optimization see: 'Active Portfolio Management based on

Error Correction Neural Networks' and 'Optimal Asset Allocation for a

Large Number of Investment Oportunities'.

 

Concerning Undershooting see: "Undershooting: Modeling Dynamical Systems

by Time Grid Refinements' and concerning causal-retro-causal networks

see: 'Prosody Generation by Causal-Retro-Causal Error Correction Neural

Networks'.

 

Concerning Neuro-Fuzzy see: 'Neuro-Fuzzy Systems for Data Analysis'.

 

The multi-agent part is best covered in the Ph.D. thesis of my colleague

Ralph Grothmann. This thesis refers also to the main parts of the

analytical sections of the lecture. The thesis can be downloaded from:

 

http://elib.suub.uni-bremen.de/publications/dissertations/E-Diss437_grothmann.pdf

 

CV Hans Georg Zimmerman, lecturer during the course :

Study of mathematics, computer science and operations research in

Bonn, diploma 1982 in mathematics. Research in applications of control

theory in economics at the University of Bonn until 1987, PhD 1987 in

economics. Since 1987 at the department for Corporate Technology,

Siemens AG. Research in circuit simulation, since 1988 in neural

networks. Current research interests: Optimization, time series analysis

and economic aplications of neural networks. Since 1990 leader of the

project group 'Complex Systems Analysis by Neural Networks'. Head of

the SENN development (Simulation Environment for Neural Networks).

Work in the development of feedforward, recurrent and neurofuzzy network

architectures and algorthms for the modeling of economical dynamical

systems.

 

Participation/Registration:

 

If you want to participate, please send your name, affiliation and e-mail address to

Jan-Erik Käck jan-erik.kack@mdh.se . Please send your registration before Sept 8, for planning reasons.

 

Everyone interested is welcome to participate. The fee will be 300 SEK (or 30 Euro) per participant and can be paid directly on site or be added to the other fee for those participating in the SIMS2003. The fee will not cover any meals.