2D Discontinuous Function Approximation with Real-Valued Grammar-Based Classifier System

摘要

Learning classifier systems (LCSs) are rule-based, evolutionary learning systems. Recently, there is a growing interest among the researchers in exploring LCSs implemented in a real-valued environment, due to its practical applications. This paper describes the use of a real-valued Grammar-based Classifier System (rGCS) in a task of 2D function approximation. rGCS is based on Grammar-based Classifier System (GCS), which was originally used to process context free grammar sentences. In this paper, we propose an extension to rGCS, called Simple Accept Radius (SAR) mechanism, that filters invalid and unexpected input real values. Performance evaluations show that the additional Simple Accept Radius mechanism enables rGCS to accurately approximate 2D discontinuous function. Performance comparisons with another real-valued LCS show that rGCS yields competitive performance.