Authors:
Dante Niewenhuis
1
and
Daan van Den Berg
2
Affiliations:
1
Master Artificial intelligence, Informatics Institute, University of Amsterdam, The Netherlands
;
2
Department of Computer Science, Vrije Universiteit Amsterdam, The Netherlands
Keyword(s):
Evolutionary Algorithms, Continuous Problems, Benchmarking, Instance Hardness.
Abstract:
This paper is an exploration into the hardness and evolvability of default benchmark test functions. Some very well-known traditional two-dimensional continuous benchmark test functions are evolutionarily modified to challenge the performance of the plant propagation algorithm (PPA), a crossoverless evolutionary method. For each traditional benchmark function, only its scalar constant parameters are mutated, but the effect on PPA’s performance is nonetheless enormous, both measured in objective deficiency and in the success rate. Thereby, a traditional benchmark functions’ hardness can thereby indeed be evolutionarily increased, and an especially interesting observation is that the evolutionary processes seem to follow one of three specific patterns: global minimum narrowing, increase in ruggedness, or concave-to-convex inversion.