Skip to main content

Research Repository

See what's under the surface

Advanced Search

The Baldwin effect hinders self-adaptation

Smith, Jim

Authors

Jim Smith James.Smith@uwe.ac.uk
Professor in Interactive Artificial Intelligence



Contributors

Thomas Bartz-Beielstein
Editor

Jurgen Branke
Editor

Bogdan Filipic
Editor

Abstract

The “end-game” of evolutionary optimisation is often largely governed by the efficiency and effectiveness of searching regions of space known to contain high quality solutions. In a traditional EA this role is done via mutation, which creates a tension with its other different role of maintaining diversity. One approach to improving the efficiency of this phase is self-adaptation of the mutation rates. This leaves the fitness landscape unchanged, but adapts the shape of the probability distribution function governing the generation of new solutions. A different approach is the incorporation of local search – so-called Memetic Algorithms. Depending on the paradigm, this approach either changes the fitness landscape (Baldwinian learning) or causes a mapping to a reduced subset of the previous fitness landscape (Lamarkian learning). This paper explores the interaction between these two mechanisms. Initial results suggest that the reduction in landscape gradients brought about by the Baldwin effect can reduce the effectiveness of self-adaptation. In contrast Lamarkian learning appears to enhance the process of self-adaptation, with very different behaviours seen on different problems.

Publication Date Sep 9, 2014
Print ISSN 0302-9743
Peer Reviewed Peer Reviewed
Volume 8672
Pages 120-129
Series Title Lecture Notes in Computer Science
Book Title Parallel Problem Solving from Nature – PPSN XIII
ISBN 9783319107615
APA6 Citation Smith, J. (2014). The Baldwin effect hinders self-adaptation. In J. Branke, B. Filipic, J. Smith, & T. Bartz-Beielstein (Eds.), Parallel Problem Solving from Nature – PPSN XIII, 120-129. Springer. https://doi.org/10.1007/978-3-319-10762-2_12
DOI https://doi.org/10.1007/978-3-319-10762-2_12
Keywords evolutionary computation, self-adaptation, memetics
Publisher URL http://dx.doi.org/10.1007/978-3-319-10762-2_12
Additional Information Title of Conference or Conference Proceedings : Parallel Problem Solving from Nature PPSN 13
;