Genetic Programming for Automatic Design of Parameter Adaptation in Dual-Population Differential Evolution

Created by W.Langdon from gp-bibliography.bib Revision:1.7644

@InProceedings{stanovov:2023:ECADA,

• author = "Vladimir Stanovov and Eugene Semenkin",
• title = "Genetic Programming for Automatic Design of Parameter Adaptation in {Dual-Population} Differential Evolution",
• booktitle = "13th Workshop on Evolutionary Computation for the Automated Design of Algorithms",
• year = "2023",
• editor = "Daniel Tauritz and John Woodward and Emma Hart",
• pages = "1736--1743",
• address = "Lisbon, Portugal",
• series = "GECCO '23",
• month = "15-19 " # jul,
• organisation = "SIGEVO",
• publisher = "Association for Computing Machinery",
• publisher_address = "New York, NY, USA",
• keywords = "genetic algorithms, genetic programming, parameter adaptation, hyper-heuristic, differential evolution",
• isbn13 = "9798400701191",
• DOI = "doi:10.1145/3583133.3596310",
• size = "8 pages",
• abstract = "The parameter adaptation is one of the main problems in many evolutionary algorithms, including differential evolution. Instead of manual development of new methods, a hyper-heuristic approach can be used, where an algorithm is applied to search for parameter adaptation scheme. In this study the symbolic regression genetic programming is applied to design parameter adaptation method for differential evolution algorithm with two populations L-NTADE. Due to algorithmic scheme different from popular L-SHADE, the L-NTADE may require specific adaptation mechanisms. Each solution in genetic programming consists of three trees, which generate scaling factor values based on current resource, success rate and current values in the memory cells, containing scaling factor and crossover rate. The training is performed on a set of 30 benchmark functions from CEC 2017 competition on numerical optimization, and at every generation of genetic programming new problem dimension, computational resource, optima location and rotation matrices are generated for every test function. The testing is performed on two benchmarks, CEC 2017 and CEC/GECCO 2022. The results comparison shows that the automatically designed parameter adaptation heuristics are capable of outperforming the success-history adaptation in many cases, including high-dimensional problems and problems with different computational resource.",
• notes = "GECCO-2023 A Recombination of the 32nd International Conference on Genetic Algorithms (ICGA) and the 28th Annual Genetic Programming Conference (GP)",

}

Genetic Programming entries for Vladimir Stanovov Eugene Semenkin

Citations