The scope of this research is Artificial Learning Methods and how they scale to similar and related problems. In particular, the newly developed suite of methods based on XCSwith Code Fragmentswill be investigated as it has shown promise in this domain. Code fragments are Genetic-Programming-like trees that encapsulate building blocks of knowledge. The usage of code fragments in the XCS system enabled the solution of previously intractable, complex, boolean problems, e.g. the 135 bit multiplexer domain. However, it was not previously possible to replace functionality at nodes with learned relationships, which restricted scaling to larger problems and related domains. The aim of this research is to reuse learned rule sets as functions. The academic thesis is that reusing learned rule sets as functions within the nodes of code fragments, as building blocks within a learning system, will result in scalability both within the problem domain and in related problem domains.

Objective one is to determine the feasibility of reusing rule sets as functions. Objective two is to determine what types of functions can be reused. A number of systems will be developed and will be tested against numerous types of problems. Objective three is to develop a mechanism for the automatic selection of functions. The functions should be adaptive to a domain. Objective four is to enable scaling of the system to similar and related domains.

Preliminary work has shown that functions can be stored along with code fragments produced for a solution to a problem. The current methods being developed seek to reuse building blocks of knowledge in the inner nodes, as well as the leaf nodes; it is anticipated that the functionality will be reused to increase the scalability.

The benefit of this approach is that problems more difficult than the 135 bit multiplexer will be solvable in a reasonable amount of time and with comparatively similar computing resources as are currently needed. The goal is to create a technique that can scale to problems in a domain larger than previously practical and transfer the learned building blocks into related domains.