Investigating the influence of survival selection and fitness estimation method in genotype-based surrogate-assisted genetic programming

This paper focuses on surrogate-assisted genetic programming (SAGP), an efficient evolutionary program optimization approach based on the surrogate estimation of the fitness value. In particular, we use a genotype-based SAGP (G-SAGP), which uses the nearest neighbor method according to the tree stru...

Full description

Saved in:
Bibliographic Details
Published in:Artificial life and robotics 2023-02, Vol.28 (1), p.181-191
Main Authors: Harada, Tomohiro, Kino, Sohei, Thawonmas, Ruck
Format: Article
Language:English
Subjects:
Artificial Intelligence
Benchmarks
Computation by Abstract Devices
Computer Science
Control
Evolutionary computation
Fitness
Genetic algorithms
Mechatronics
Optimization
Original Article
Regression
Robotics
Survival
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper focuses on surrogate-assisted genetic programming (SAGP), an efficient evolutionary program optimization approach based on the surrogate estimation of the fitness value. In particular, we use a genotype-based SAGP (G-SAGP), which uses the nearest neighbor method according to the tree structure similarity as the surrogate evaluation. This article investigates the influence of the survival selection and the fitness estimation method in G-SAGP to improve its performance. For the survival selection, we compare the ( μ , λ ) selection and the ( μ + λ ) selection, which are both commonly used in evolutionary computation methods. On the other hand, the conventional G-SAGP uses the nearest neighbor method for the fitness estimation method, while this article attempts the k -nearest weighted average regression. We conduct experiments using symbolic regression problems, which are usually used as the GP benchmark. The experiments compare different survival selection methods and different fitness estimation methods. The experimental results show that G-SAGP using the ( μ + λ ) selection can reduce the number of generations while maintaining a higher success ratio. In addition, for the fitness estimation method, the nearest neighbor regression is enough to achieve a high success ratio with a smaller number of generations. In contrast, the k -nearest weighted average regression with a large k can perform better in some benchmark problems.
ISSN:1433-5298
1614-7456
DOI:10.1007/s10015-022-00821-3