Human-inspired Scaling in Learning Classifier Systems: Case Study on the n-bit Multiplexer Problem Set

Learning classifier systems (LCSs) originated from artificial cognitive systems research, but migrated such that LCS became powerful classification techniques. Modern LCSs can be used to extract building blocks of knowledge in order to solve more difficult problems in the same or a related domain. The past work showed that the reuse of knowledge through the adoption of code fragments, GP-like sub-trees, into the XCS learning classifier system framework could provide advances in scaling. However, unless the pattern underlying the complete domain can be described by the selected LCS representation of the problem, a limit of scaling will eventually be reached. This is due to LCSs' 'divide and conquer' approach utilizing rule-based solutions, which entails an increasing number of rules (subclauses) to describe a problem as it scales. Inspired by human problem solving abilities, the novel work in this paper seeks to reuse learned knowledge and learned functionality to scale to complex problems by transferring them from simpler problems. Progress is demonstrated on the benchmark Multiplexer (Mux) domain, albeit the developed approach is applicable to other scalable domains. The fundamental axioms necessary for learning are proposed. The methods for transfer learning in LCSs are developed. Also, learning is recast as a decomposition into a series of sub-problems. Results show that from a conventional tabula rasa, with only a vague notion of what subordinate problems might be relevant, it is possible to learn a general solution to any n-bit Mux problem for the first time. This is verified by tests on the 264, 521 and 1034 bit Mux problems.