Sharpness-Aware Minimization in Genetic Programming

Created by W.Langdon from gp-bibliography.bib Revision:1.8311

@InProceedings{Bakurov:2024:GPTP,

• author = "Illya Bakurov and Nathan Haut and Wolfgang Banzhaf",
• title = "Sharpness-Aware Minimization in Genetic Programming",
• booktitle = "Genetic Programming Theory and Practice XXI",
• year = "2024",
• editor = "Stephan M. Winkler and Wolfgang Banzhaf and Ting Hu and Alexander Lalejini",
• series = "Genetic and Evolutionary Computation",
• pages = "151--175",
• address = "University of Michigan, USA",
• month = jun # " 6-8",
• publisher = "Springer",
• keywords = "genetic algorithms, genetic programming, SAM",
• isbn13 = "978-981-96-0076-2",
• URL = "https://arxiv.org/abs/2405.10267",
• DOI = "doi:10.1007/978-981-96-0077-9_8",
• size = "22 pages",
• abstract = "Sharpness-Aware Minimization (SAM) was recently introduced as a regularization procedure for training deep neural networks. It simultaneously minimizes the fitness (or loss) function and the so-called fitness sharpness. The latter serves as a measure of the nonlinear behaviour of a solution and does so by finding solutions that lie in neighborhoods having uniformly similar loss values across all fitness cases. we adapt SAM for tree Genetic Programming (TGP) by exploring the semantic neighborhoods of solutions using two simple approaches. By capitalising upon perturbing input and output of program trees, sharpness can be estimated and used as a second optimization criterion during the evolution. To better understand the impact of this variant of SAM on TGP, we collect numerous indicators of the evolutionary process, including generalization ability, complexity, diversity, and a recently proposed genotype-phenotype mapping to study the amount of redundancy in trees. The experimental results demonstrate that using any of the two proposed SAM adaptations in TGP allows (i) a significant reduction of tree sizes in the population and (ii) a decrease in redundancy of the trees. When assessed on real-world benchmarks, the generalization ability of the elite solutions does not deteriorate.",
• notes = "Published in 2025 after the workshop",

}

Genetic Programming entries for Illya Bakurov Nathaniel Haut Wolfgang Banzhaf

Citations