Abstract
This paper considers some issues related to the apportionment of credit problem in Genetic Based Machine Learning systems (GBML). A GBML system is composed of three major subsystems. The first one, the performance subsystem, is a parallel adaptive rule-based system where the knowledge base is a set of rules expressed in a low-level syntax. The second subsystem, called Genetic Algorithm (GA), is a procedure that searches in the rule space by means of genetic operators modelled according to natural genetic operators (e.g. reproduction, crossover, mutation). The third subsystem faces the apportionment of credit problem, i.e. how to evaluate the quality of existing rules. In this paper we propose an apportionment of credit algorithm, called Message-Based Bucket Brigade, in which messages instead of rules are evaluated. A rule quality is then a function of the value of the messages matching the rule conditions, of the rule conditions specificity and of the value of the message the rule tries to post. This approach gives a solution to the default hierarchy formation problem, i.e. the problem of creating set of rules in which default rules cover broad categories of system responses, while specific ones cover situations in which default rules are incorrect. A comparison with other approaches to default hierarchy formation is presented. The final section presents conclusions and suggests directions for further research.
Chapter PDF
Keywords
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
References
Goldberg D.E. & Wilson S.W., "A Critical Review of Classier Systems", Proceedings of the Third International Conference on Genetic Algorithms, June 4–7 1989, Morgan Kaufmann (pp.244–255).
Holland J.H., "Adaptation in natural and artificial systems", Ann Arbor: The University of Michigan Press, 1975.
Riolo R.L., "Bucket Brigade Performance: II. Default Hierarchies" Proceedings of the Second International Conference on Genetic Algorithms, July 28–31 1987, Lawrence Erlbaum (pp.196–201).
Wilson S.W., "Bid Competition and Specificity Reconsidered", Complex Systems, 2(6), 1988, pp. 705–723.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1991 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Dorigo, M. (1991). Message-based bucket brigade: An algorithm for the apportionment of credit problem. In: Kodratoff, Y. (eds) Machine Learning — EWSL-91. EWSL 1991. Lecture Notes in Computer Science, vol 482. Springer, Berlin, Heidelberg. https://doi.org/10.1007/BFb0017018
Download citation
DOI: https://doi.org/10.1007/BFb0017018
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-53816-5
Online ISBN: 978-3-540-46308-5
eBook Packages: Springer Book Archive