Theory and application of learning automata.

Although the theoretical performance of many learning automata has been considered, the practical operation of these automata has received far less attention. This work starts with the construction of two action Tsetlin and Krylov automata. The performance of these automata has been measured in stationary and non-stationary environments. The operation of a hierarchical automaton controlling the memory size of a Tsetlin automaton is also investigated. Two new automata are proposed with the aim of avoiding the operational disadvantages of the Tsetlin automaton. These automata have been tested using a computer simulation and, in addition, theoretical performance results have been calculated and compared with results for Tsetlin, Krylov and Lri automata. A model of a non-autonomous environment is simulated and its operation analysed theoretically. A more accurate model is analysed, and its operation with a Lri automaton examined and compared to theoretical predictions. The requirements for learning automata to operate successfully in non-autonomous environments are considered and it is shown that the Lrp and Lri automata do not converge to the optimum for a non-autonomous environment. Three automata are proposed, which are designed to operate in non-autonomous environments. Their performances are compared to those of the Lrp and Lri automata. The operation of automata in a hierarchical learning system and in cooperative and competitive games is considered. In these situations the performance of the new automata is compared to that of the Lrp and Lri automata. Finally, two applications of learning automata are investigated. The first considers the Tsetlin allocation scheme, gives a modification that increases the performance and makes a comparison with a scheme using other learning automata. The second involves the selection of a processor in a multiprocessor computer system and compares a scheme using learning automata with a fixed scheduling discipline.

Name dc:type.qualificationname

PhD

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

Robert Gordon's Institute of Technology

Year dc:date.issued

1980

Author dc:creator

• Mackie, Neil James

Advisor dc:contributor.advisor

• P. Mars and N. Deans

• Machine learning
• Artificial intelligence
• Learning automata

Language dc:language

en

Identifier

oai:rgu-repository.worktribe.com:1993304
https://doi.org/10.48526/rgu-wt-1993304

OAI identifier oai:identifier

oai:rgu-repository.worktribe.com:1993304