Feature Representations in Connectionist Systems
- 177 Downloads
This paper has two goals: to demonstrate the importance of the feature representation chosen for a connectionist model, and to examine the properties of a particular model, the back-propagation algorithm, in conditions intended to simulate the ‘graceful degradation’ encountered in human ageing. In a concept learning simulation, the number of zero or near-zero values of the input features and in the responses of the hidden units was found to influence speed of learning, strength of response to a prototype, and performance with distorted input (the latter two being inversely related). Degradation of the network prior to learning enhanced prototype performance but disrupted distortion performance. In the light of these results we discuss the design of efficient learning algorithms and the potentiality of these networks as models for human ageing.
Unable to display preview. Download preview PDF.
- Burke, D. M., & Harrold, R. M. (1988). Automatic and effortful processes in old age: Experimental and naturalistic processes. In L. Light & D.M. Burke (Eds.) Language, memory and aging (pp. 100–116). Cambridge: Cambridge University Press.Google Scholar
- Chomsky, N., & Halle, M. (1968). The sound pattern of English. New York: Harper & Row.Google Scholar
- Greenberg, J. (1966). Language universals. The Hague: Mouton.Google Scholar
- Howard, D.V. (1988) Aging and memory activation: The priming of semantic and episodic memories. In L. Light & D.M. Burke (Eds.) Language, Memory and Aging (pp. 77–79). Cambridge: Cambridge University Press.Google Scholar
- McClelland, J. L., & Rumelhart, D. E. (1988). Explorations in parallel distributed processing. Cambridge, MA: MIT Press.Google Scholar
- Trubetzkoy, N. S. (1939/1969). Grundzüge der Phonologie [Principles of phonology]. Los Angeles, CA: University of California Press.Google Scholar