Landscape Analysis for Surrogate Models in the Evolutionary Black-Box Context

Authors: Z. Pitra, J. Tumpach, Martin Holeňa
Publication year: 2025
Publication type: Konferenční příspěvek (mezinárodní konf.)
Periodical: GECCO '25 Companion: Proceedings of the Genetic and Evolutionary Computation Conference Companion
Bibliographic citation: PITRA, Z., TUMPACH, J., HOLEŇA, Martin. Landscape Analysis for Surrogate Models in the Evolutionary Black-Box Context. In: OCHOA, G., FILIPIČ, B., eds. GECCO '25 Companion: Proceedings of the Genetic and Evolutionary Computation Conference Companion. New York: Association for Computing Machinery (ACM), 2025, S. 51-52. ISBN 979-8-4007-1464-1
Dostupné z: doi:10.1145/3712255.3734236.
DOI: 10.1145/3712255.3734236
ID: 641201
Online record: View online record

Abstract

This paper, originally published in Evolutionary Computation Journal [11], investigates the interplay between surrogate model performance, model settings, and black-box landscape features within the context of Covariance Matrix Adaptation Evolution Strategy (CMA-ES). Our focus is on understanding how landscape characteristics influence surrogate model accuracy during evolutionary optimization, aiming to inform the automated selection and tuning of surrogate models. We perform a comprehensive feature analysis, identifying robust and informative landscape features relevant to surrogate modeling. The analysis explores the error dependencies of four models across 39 settings, utilizing three methods for input data selection, drawn from surrogate-assisted CMA-ES runs on noiseless benchmarks within the Comparing Continuous Optimizers framework. The insights gained can help in the development of adaptive surrogate modeling strategies and metalearning approaches for evolutionary computation.