Knowledge representation with neural networks (NGN.4480)

The aims of this project are to develop novel theory, methods and implementations for learning and reasoning in a complex dynamic multi-sensory environment. The approach to reasoning and learning is based on the axioms of probability theory. It is argued that such an approach is the most attractive way to design systems for reasoning and learning that are capable of reliable and robust performance in complex real-world environments. The research aims at design of algorithms to enable learning and reasoning involving up to the order of 1000 variables. This allows for applications which are order of magnitude larger than currently possible.

This project addresses essential aspects of the probabilistic treatment of uncertain reasoning and learning. Specific issues are

1. Learning and inference in probabilistic knowledge representations. The time required for learning and inference is in general too large for practical applications but special architectures for which fast algorithms can be derived will be designed.
2. The study of non-equilibrium dynamics as required for parallel implementation of the stochastic networks.
3. Rule extraction from trained networks and
4. integration of domain knowledge and data.

In addition, the strength of the methods will be demonstrated in this project in several demonstrators and applications in collaboration with industry.

The results of this project will consist of general methods for learning and reasoning in complex real world environments. These methods are of crucial importance for large scale applications in self-organizing information databases, human-machine dialogue systems, reasoning in large knowledge domains, pattern recognition and robotics.

Dr. H.J. Kappen
Stichting Neurale Netwerken
Lab. Med. en Biofysica
Postbus 9101
6500 HB Nijmegen