Affiliations: [a] School of ITEE, University of New South Wales, Australian Defence Force Academy, Canberra, Australia | [b] School of Computer Science & Engineering, College of Engineering, Seoul National University, Korea
Abstract: We build on our previous feasibility studies [18,20], which demonstrated the impact of evaluation during development in the DEVTAG system, and here present a full-fledged developmental system – Developmental Tree Adjoining Grammar Guided Genetic Programming (DTAG3P) with developmental evaluation, based on Tree-Adjoining Grammars (TAG). While DEVTAG used only a trivial developmental process, DTAG3P uses L-systems to encode TAG derivation trees, because the L-systems permit a full developmental process. DEVTAG was previously shown to dramatically out-perform standard Genetic Programming (GP) on some structured families of problems; here, we examine DTAG3P's performance on these families, and find a further major increment in performance over DEVTAG. DTAG3P achieves this despite dispensing with two extra control parameters which were necessary with DEVTAG.
