scientific article; zbMATH DE number 1154419

From MaRDI portal
Publication:4389727





Starting point of this thesis is the theory of anticipatory behaviour control by Hoffmann (1992). Hoffmann formulates a learning mechanism based on S-R-C(ant)-relationships. S is the situation, R a reaction and C(ant) means the anticipatory behaviour consequences. The author discusses the following questions: -- Is the learning mechanism by Hoffmann a complete theory? -- Are there any contradictions? -- In which parts there are open points? The main purpose of investigation is to develop a learning algorithm based on the learning mechanism by Hoffmann. Section 1 of the thesis contains an introduction to the theory of anticipatory behaviour control. It is seen as beyond classic amplifying learning. Section 2 is an introduction to classifier systems. The essential components input interface, output interface, message list and classifier list are explained. After that follows a formal definition of classifier systems. The different concepts of classifier list are investigated and compared, too. The results are a new definition of classifiers and rules to encoding of input and output interfaces. Section 3 is the discussion of anticipatory classifier systems (acs). The predicate match and the function passthrough is explained extensively. The argue chain is component of acs, behaviour of acs, learning within acs. The result is the development of a learning algorithm. Section 4 comprises an evaluation of anticipatory classifier systems. Technical application is discussed. In section 5 are pointed out borders and possibilities of extensions. So the author thinks that in the next time acs will be able to recognize environmental situations by means of discrete values. Changes of learning algorithm are needed for that.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4389727)