The first Dagstuhl Seminar on Neural Computing had been organized by Wolfgang Maass (Technische Univrsität Graz, Austria), Christoph von der Malsburg (Ruhr Universität Bochum), Eduardo Sontag (Rutgers University, USA) und Ingo Wegener (Universität Dortmund). It brought together 35 participants from 6 countries, among these 6 from overseas.

The seminar consisted of 28 plenary talks, 5 tutorials, a panel discussion, an open problem session, and numerous discussions as well as two extra talks in smaller groups. The panel discussion covered a number of common problems in theoretical and experimental research on neural networks, such as the proper choice of models and benchmark problems for the investigation of computing and learning on biological and artificial neural networks. Tutorials were given by Christoph von der Malsburg (”The binding problem of neural networks”), Wulfram Gerstner (”Models of spiking neurons”), Angus Macintyre (”The VC-dimension of neural networks”), Jehoshua Bruck/Thomas Hofmeister/Matthias Krause (”The computational complexity of threshold circuits”), and Peter Bartlett (”Learning on neural nets from the point of view of computational learning theory”).

A thorough understanding of neural networks requires methods and results from quite diverse disciplines such as computer science, biology, engineering, physics, statistics, and mathematics. Therefore we had invited to this Dagstuhl seminar besides computer scientists also experts for neural networks from these other disciplines. The five tutorials provided a substantial common knowledge background and sufficient mutual ”language-understanding”, so that the participants could follow the presentations of new research results from all of the here represented disciplines. Many interesting interdisciplinary discussions ensued, especially since many of the participants never had the opportunity before to attend a meeting of this type.

Apart from this interdisciplinary aspects, the meeting also provided a forum for the presentation of a number of exciting new research results about neural networks. As just one example one could mention the plenary talk, and two evening sessions with technical details, which were given by Angus Macintyre from Oxford University. He reported on the solution of a well-known open problem in theoretical research on neural networks, which he had recently achieved jointly with Marek Karpinski from the University of Bonn. Their seminal result, which was made public for the first time at this Dagstuhl meeting, provides a polynominal upper bound for the Vapnik-Chervonenkis dimension of sigmoidal neural nets (previously only a double-exponential upper bound had been achieved).

This new result, as well as some other new research results which were presented at this seminar, will be published in a special issue of the Journal of Computer and System Sciences for the first Dagstuhl Seminar on Neural Computing.

Creative Commons BY 3.0 Unported license