VCS: variable classifier systems
Lingyan Shu, Jonathan Schaeffer · 1989
Classifier systems (CS) have proven to be useful tools for the study of genetic algorithm based learning. Unfortunately, there are a number of difficulties with the formalization that limit the representational capabilities and, hence, its problem solving abilities and the speed at which it can learn. This paper introduces VCS - Variable Classifier Systems - that augment the traditional CS with the binding of constants in messages to variables in rule conditions. For a large class of problems, VCS allows for a more succinct representation of the solution space than is possible with CS, increasing the likelihood of a genetic search successfully solving the problem. Variables make it possible for these systems to represent information in ways similar to high level symbolic representations, narrowing the gap between classifier systems and conventional learning systems.