Authors:
Fu Xing Long
1
;
Diederick Vermetten
2
;
Anna Kononova
2
;
Roman Kalkreuth
3
;
Kaifeng Yang
4
;
Thomas Bäck
2
and
Niki van Stein
2
Affiliations:
1
BMW Group, Knorrstraße 147, Munich, Germany
;
2
LIACS, Leiden University, Niels Bohrweg 1, Leiden, The Netherlands
;
3
Computer Lab of Paris 6, Sorbonne Université, Paris, France
;
4
University of Applied Sciences Upper Austria, Softwarepark 11, 4232, Hagenberg, Austria
Keyword(s):
Function Generator, Genetic Programming, Exploratory Landscape Analysis, Instance Spaces.
Abstract:
Within the optimization community, the question of how to generate new optimization problems has been gaining traction in recent years. Within topics such as instance space analysis (ISA), the generation of new problems can provide new benchmarks which are not yet explored in existing research. Beyond that, this function generation can also be exploited for solving expensive real-world optimization problems. By generating fast-to-evaluate functions with similar optimization properties to the target problems, we can create a test set for algorithm selection and configuration purposes. However, the generation of functions with specific target properties remains challenging. While features exist to capture low-level landscape properties, they might not always capture the intended high-level features. We show that it is challenging to find satisfying functions through a genetic programming (GP) approach guided by the exploratory landscape analysis (ELA) properties. Our results suggest th
at careful considerations of the weighting of ELA properties, as well as the distance measure used, might be required to evolve functions that are sufficiently representative to the target landscape.
(More)