Wichert, Andreas Miroslaus
FacultiesFakultät für Informatik
LicenseStandard (Fassung vom 03.05.2003)
Currently neural networks are used in many different domains. But are neural networks also suitable for modeling problem solving, a domain which is traditionally reserved for the symbolic approach? This central question of cognitive science is answered in this work. It is affirmed by corresponding neural network models. The models have the same behavior as the symbolic models. However, also additional properties resulting from the distributed representation emerge. It is shown by comparison of those additional abilities with the basic behavior of the model that the additional properties lead to a significant algorithmic improvement. This is verified by statistical hypothesis testing. The associative computer, a neural model for a reaction system based on the assembly theory, is introduced. It is shown that planning can be realized by a neural architecture that does not use symbolic representation. A crucial point is the description of states by pictures. The human ability to process images and understand what they mean in order to solve a problem holds an important clue to how the human thought process works. This clue is examined by empirical experiments with the associative computer. One general conclusion from the experiments is the claim that it is possible to use systematically associative structures to perform reasoning by forming chains of associations. In addition, beside symbolical problem solving, pictorial problem solving is possible.
Subject HeadingsNeural networks: Computer science [LCSH]
Problem solving: Artificial intelligence [LCSH]
Man-machine systems [MeSH]