Abstract
Approaches for rule based control often rely heavily on the pre-classification of the state space for their success. In the pre-determined regions individual or groups of rules may be learned. Clearly, the success of such strategies depends on the quality of the partitioning of the state space. When no such a priori partitioning is available it is a significantly more difficult task to learn an appropriate division of the state space as well as the associated rules. Yet another layer of potential difficulty is the nature of the reinforcement applied to the rules since it is not always possible to generate an immediate reinforcement signal to supply judgement on the efficacy of activated rules. One approach to combine the joint goals of determining partitioning of the state space and discovery of associated appropriate rules is to use a genetic algorithm employing a restricted mating policy to generate rule clusters which dominate regions of the state space thereby effecting the required partioning. Such rale clusters are termed niches. A niching algorithm, which includes a ‘creche’ facility to protect ‘inexperienced’ classifiers, and the results of determining a simple 2D state space using an immediate reward scheme are presented. Details of how the algorithm may modified to incorporate a delayed reinforcement scheme on a real-world beam balancing control problem are reported.
Preview
Unable to display preview. Download preview PDF.
References
Booker L.B. 1982. Intelligent Behaviour as an Adaptation to the Task Environment.Doctoral Thesis, University of Michigan. Dissertation Abstracts International, 43(2),469B
Davidor Y. 1991. A Naturally OccurringNiche and Specied Phenomenon. In Proc. 4th. Int. Conf. on Genetic Algorithms pp 257–263 Ed: Belew R. and Booker L.
Fogarty T.C. 1989. Learning New Rules and Adapting Old Ones with the Genetic Algorithm. In Artificial Intelligence In Manufacturing. Ed: G.Rzevski. Springer Verlag pp275–290
Fogarty T.C. 1994. Selectionist Reinforcement in Learning Classifier Systems. BTC-CSM Technical Report University of the West of England, Bristol.
Huang R. 1993. Evolving Prototype Rules for Process Control. Ph.D. thesis. University of the West of England, Bristol.
Michie D. and Chambers R.A. 1968. BOXES: An Experiment in Adaptive Control. In Machine Intelligence 2.
Moore A. 1991. Variable Resolution dynamic Programming: Efficiently Learning Action Maps in Multivariate Real-valued State-spaces. In Machine Learning: Proc. 8th Int. Workshop. Ed: Birnbaum L. & Collins G. Pub: Morgan Kaufmann.
Odetayo M.O. and McGregor D.R. 1989. Genetic Algorithm for Inducing Control Rules for a Dynamic System. Proc 3rd Int. Conf. on Genetic Algorithms pp 177–182.
Renders J-M., Nordvik J-P. and Bersini H. 1992. Genetic Algorithms for Process Control: A Survey. In Proc. of International Symposium on Artificial Intelligence In Real-Time Control.
Rosen B. and Goodwin J. 1993. Dynamic Flight Control with Adaptive Coarse Coding. In Proc. 2nd Int. Conf. on Simulation of Adaptive Behaviour. MIT Press.
Twardowski K. 1993. Credit Assignment for Pole Balancing with Learning Classifier Systems. In Proc. 5th. Int. Conf. on Genetic Algorithms.
Valenzuela-Rendon M. 1991. The Fuzzy Classifier System: A classifier System for Continuously Varying Variables. In Proc. 4th. Int. Conf. on Genetic Algorithms pp346–353.
Woodcock N., Hallam N. and Picton P. 1991. Fuzzy Boxes as an Alternative to Neural Networks for Difficult Control Problems. Proc. 6th Int. Conf on AI in Engineering.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1994 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Melhuish, C., Fogarty, T.C. (1994). Applying a restricted mating policy to determine state space niches using immediate and delayed reinforcement. In: Fogarty, T.C. (eds) Evolutionary Computing. AISB EC 1994. Lecture Notes in Computer Science, vol 865. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-58483-8_17
Download citation
DOI: https://doi.org/10.1007/3-540-58483-8_17
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-58483-4
Online ISBN: 978-3-540-48999-3
eBook Packages: Springer Book Archive