Distributed Representations in Radial Basis Function Networks
This chapter locates Gaussian radial basis function (RBF) networks within the school of connectionist modelling that has successfully exploited the properties of multi layer perceptrons (MLPs). In particular, the highly regarded use of distributed representations in MLPs is established in RBF networks. RBF networks have been used to learn a categorization task, and the way in which the networks have learned the task is interpreted in terms of localist and distributed representations. The problem of catastrophic forgetting as a psychologically implausible feature of MLPs, and an undesirable property of neural network engineering systems, is also reconsidered.
KeywordsRadial Basis Function Receptive Field Radial Basis Function Network Hide Unit Training Pattern
Unable to display preview. Download preview PDF.
- 1.Rumelhart DE, McClelland JL: On learning the past tenses of English verbs. In Parallel Distributed Processing: Explorations in the Microstructure of Cognition, Vol. 2, Rumelhart DE, McClelland JL (Eds). MIT Press, Cambridge, Mass, 1986, pp 216–271Google Scholar
- 6.Bishop CM. Neural networks for pattern recognition. Clarendon Press, Oxford, 1995Google Scholar
- 11.French RM. Using semi-distributed representations to overcome catastrophic forgetting in connectionist networks. CRCC TR 51–1991, 1991Google Scholar
- 12.Clark A. Associative Engines: Connectionism, Concepts, and Representational Change. MIT Press, Cambridge, Mass, 1993Google Scholar