Connectionism

from Wikipedia, the free encyclopedia

The connectionism is a problem-solving approach to cybernetics and deals with the behavior of networked systems based on combinations of artificial information processing units. Behavior is understood as the product of a large number of interacting components that mutually influence one another. With the aid of artificial neural networks consisting of an apparent is chaos arising system order simulated. Application areas of connectionism include neurophysiology , psychology , biology , linguistics , neuroinformatics ,Movement Science and Artificial Intelligence Research.

Problem solving with connectionist systems

Problem solving always consists of the following steps, regardless of the respective fields of application:

  • Collect information
  • Form a model
  • Create forecast
  • Check the result

The step of modeling is undoubtedly the most difficult. Expert systems , simulations and numerical calculations require detailed knowledge of the system that is to be examined. Their constructivist approach is based on the hypothesis that systems can be completely symbolically described and thus algorithmized by breaking them down into subsystems of a certain structure.

In a connectionist model, the attempt is made to reproduce the (external) behavior of a system as a whole through the connection of a large number of relatively simple and often very similar units that are connected to one another in a dense network. These units work locally and only communicate with others via signals over links.

A connectionist model system is set up for selected examples of the system to be investigated in such a way that it shows the same behavior as its model under the same conditions. In these cases there is an isomorphism of behavior, the connectionist model system responds to inputs with the same outputs as its real model. Since the system behavior is not algorithmized, it is not understandable how the connectionist model system works internally, its results always arise from the interaction of all elements. The connectionist model system does not necessarily have to be isomorphic to the object of investigation. According to Smolensky, knowledge is represented sub-symbolically.

Sub-symbolic hypothesis

The derivation of knowledge arises from the interaction of a large number of units. This interaction does not allow an exact description on a conceptual level, but has to be realized directly by model processors. The model conception of a connectionist system is fundamental and independent of a concrete implementation. In addition to the well-known artificial neural networks , Frederic Vester's sensitivity model should be mentioned as an implementation of a connectionist view.

Advantages of connectionist architectures

The most important advantages of systems with connectionist architecture (example: the human brain) are:

  1. Since they don't work according to given rules, they are very adaptable .
  2. You can learn - but it takes a long time to prepare for the system to be operational.
  3. They work excellently even in incomplete data and noisy environments, a well-known example being face recognition .
  4. Due to their redundancy, they are robust if parts of the system fail.

See also

literature

Web links

Individual evidence

  1. ↑ Philip T. Quinlan: Connectionism and psychology: a psychological perspective on new connectionist researc . Harvester Wheatsheaf, New York 1991, ISBN 0-7450-0835-6 , pp. 1 .
  2. ↑ David E. Rumelhart, James L. McClelland, San Diego. PDP Research Group. University of California: Parallel distributed processing: explorations in the microstructure of cognitio . MIT Press, Cambridge, Mass. 1986, ISBN 0-262-18120-7 , pp. 76 .
  3. ^ Rainer Wollny: Movement Science: A textbook in 12 lessons. 2nd Edition. Meyer & Meyer, Aachen 2010, ISBN 978-3898991834 , p. 32.
  4. ^ Allen Newell & Herbert A. Simon : Physical Symbol System Hypothesis. 1976.
  5. ↑ Joachim Funke : Problem-solving thinking , Kohlhammer Verlag 2003, ISBN 3-17-017425-8 .