Learning by reinforcement: A psychobiological model
The current connectionist models of the learning by reinforcement paradigm make use of the delta rule to back-propagate the error. The work we present proposes a biologically inspired learning by reinforcement method. It uses only biological concepts to learn the desired outputs, as chemical substances and homeostatic regulation. On the other hand, the formulae that rule their dynamics are made with respect to the constraints imposed by the observed phenomena in behaviorist experiments of operant and classical conditioning. The authors propose that an input-output map can be expressed as a combination of these phenomena, in the sense that the task of teaching a function is to make the network to adapt itself to verify a set of phenomena between its inputs and outputs.
Unable to display preview. Download preview PDF.
- Barto, A.G., Sutton, R.S., & Watkins, C.J.C.H. (1990). Learning and sequential decision making. In: M. Gabriel & J.W. Moore (Eds.), Learning and computational neuroscience. MIT Press.Google Scholar
- Dickinson, A. (1980). Contemporary animal learning theory. Cambridge University Press.Google Scholar
- Mackintosh, N. (1983). Conditioning and associative learning. Oxford University Press.Google Scholar
- Skinner, B. F. (1938). The behavior of organism. Appleton: New York.Google Scholar
- Skinner, B. F. (1974). About behaviorism. Knopf: New York.Google Scholar
- Thorndike, E. L. (1911). Animal intelligence. McMillan: New YorkGoogle Scholar
- Vico, F.J., Sandoval, F. & Almaraz, I. (1992). A learning by reinforcement model to predict associative and non associative phenomena. In: Proceedings of the Fifth International Symposium on Knowledge Engineering, November, Seville (Spain), pp. 190–194.Google Scholar
- Watkins, C.J.C.H. (1989). Learning from delayed rewards. Ph.D. Thesis, King's College, Cambridge.Google Scholar