Dynamics is a major feature of neural modelling. The study of neural networks viewed as dynamical systems evolving under environmental pressure is an attractive perspective to overcome the limitations of classical artificial neural networks especially in the fields of vision and control. Notions such as attractors, non-stationary processes, the binding problem, dynamic programming and decision processes come from different disciplines such as dynamical systems theory, statistical physics, neurobiology, information theory and adaptive control. The DYNN workshop is a place where neurobiologists, theoreticians and engineers can exchange ideas and transfer knowledge.
The first international workshop DYNN'96 took place in ONERA in Toulouse, France, and was the occasion to compare theoretical studies and models of visual cortex with more general perceptual devices. The second workshop VI'DYNN'98 in Stockholm focused on Pulse Coupled Neural Networks and was dedicated to more applied tasks of image processing and computer vision. In DYNN'2000 all these themes were kept and new and new concepts from statistical physics were introduced such as physics "Self-organized Criticality" which describes the spatio-temporal self-organization of coupled automata. DYNN'2000 was organized at the ZiF in Bielefeld University from 11/20/2000 to 11/24/2000 in the frame of the ZiF:Resaerch Group The Sciences of Complexity: From Mathematics to Technology to a Sustainable World.
The objectives of the ZiF, scientific quality, interdisciplinarity of the research themes fitted closely the objectives of the organizers of DYNN. The support of the ZiF allowed a lot of young researchers and doctorate students to present their work and to acquire an overview of their subject by international specialists.
22 researchers were invited by the ZiF (8 members of the program committee and 14 invited lecturers). A call for papers allowed the selection of 21 communications from 7 countries. There has been about 60 participants. Among them about 10 researchers who were staying at ZiF in the framework of the research program on complexity.
The workshop was organized on the basis of five days. Some special topics were presented in the opening session about Complexity, Information theoretic approach to neural coding and modelling of meaning in language. The subsequent sessions were devoted to Dynamics and Self-Organization (I, II and III), Mean-field theory, Mathematical modelling of nervous systems, Spiking neurons (I and II), Control (I and II) and Vision.
If we compare the situation of research in neural networks ten years ago and now, it is clear that the big rendezvous between statistical artificial intelligence and neural models is past now (or future). The community of Artificial Neural Networks has turned towards statistical learning and model selection and the community of Neural Computing is now investigating time coding which is much more difficult to analyse mathematically than rate coding. Researchers has to build new tools or to adapt them from mathematics (dynamical systems and probability) and statistical physics. In the same time, biologically plausible neural networks are still a source of inspiration for engineers who deal with complex problems of control and vision with real time constraints which are still out of classical models reach. DYNN'2000 was an original meeting where these issues were discussed. Exchange between researcher from various fields will certainly be the source of new progress in the close future.
The study of self-organization of neural networks viewed as dynamical systems evolving under environmental pressure and its consequences for information processing is a growing field attracting new researchers who try to overcome the limitations of classical artificial neural networks. The connectionist framework offers a good basis for the study of complex computational phenomena. It is indeed of crucial interest to take into account both spatial and temporal aspects of information flow when one wants to study the properties of distributed systems where every local component participates to the global process. Notions such as basins of attraction, non-stationary processes, the binding problem, dynamic programming and decision processes come from different disciplines such as dynamical systems theory, statistical physics, neurobiology, information theory and adaptive control. It is important for these different communities to bring together their findings, in both perspectives of physical/biological modelling and engineering applications.
Participants Daan Archer (Amsterdam), Sonya Bahar (St. Louis, MO), Maryse Beguin (Saint Martin d’Heres), Mohamed Ben Haddou (Brüssel), Daan van den Berg (Amsterdam), Hugues Bersini (Brüssel), Dmitri Bibitchkov (Göttingen), Sander Bohte (Amsterdam), Frank Brand (München), Hilary Buxton (Brighton), Xavier Clastres (Toulouse), Emmanuel Daucé (Toulouse), Andreas Degenhard (Sutton), Valeria Del Prete (Triest), Antonio Dente (Lissabon), Christian Eurich (Bremen), Daniel Gandolfo (Marseille), Ruiyao Gao (Dublin), Wulfram Gerstner (Lausanne), Christelle Godin (Grenoble), Michael Herrmann (Göttingen), Kazushi Ikeda (Kioto), Bert Kappen (Nijmegen), Toshiki Kindo (Tokio), Jason Kinser (Manassas, VA), Camilo La Rota (La Tronche), Artur Luczak (Krakau), João Francisco Alves Martins (Setubal), Patrick McGuire (Bielefeld), Klaus Obermayer (Berlin), Kachayar Pakdaman (Paris), Laurent Perrinet (Toulouse), Armando Pires (Setubal), Mathias Quoy (Cergy-Pontoise), Stefan Reimann (Sankt Augustin), Anis Rezgui (Bielefeld), Sebastian Risau-Gusman (Grenoble), Georges Schutz (Esch Alzette), Pierre Sener (Brüssel), Madeleine Ssirugue-Collin (Marseille), Nikos Skantzos (London), Pierre-Edouard Sottas (Lausanne), Axel Steinhage (Bochum), Tanja Steinhoff (London), Ruedi Stoop (Zürich), Sebastiano Stramaglia (Bari), Jun Tani (Tokio), James Vaccaro (Rom), Bernard Victorri (Montrouge), Rui Vilela Mendes (Lissabon), Takashi Watanabe (Saitama), Torsten Wirsig (Bielefeld), Tadashi Yamazaki (Tokio)