Investigating scaling of an abstracted LCS utilising ternary and s-expression alphabets

Abstract

Utilising the expressive power of S-Expressions in Learning Classifier Systems often prohibitively increases the search space due to increased flexibility over the ternary alphabet. Selection of appropriate S-Expressions functions through domain knowledge improves scaling, as expected. Considering the Cognitive Systems roots, abstraction was included in LCS - episodic learning generalises prior to abstraction for semantic learning. This novel method is shown to provide compact results (135-MUX) and exhibits potential for scaling well (1034-MUX).