When Data Transformations Mislead Symbolic Regression: Deceptive Search Spaces in System Identification

Created by W.Langdon from gp-bibliography.bib Revision:1.8506

@InProceedings{tonda:2025:GECCOcomp2,

• author = "Alberto Tonda and Hengzhe Zhang and Qi Chen and Bing Xue and Mengjie Zhang and Evelyne Lutton",
• title = "When Data Transformations Mislead Symbolic Regression: Deceptive Search Spaces in System Identification",
• booktitle = "Symbolic Regression",
• year = "2025",
• editor = "Gabriel Kronberger and Fabricio {Olivetti de Franca} William {La Cava} and Steven Gustafson",
• pages = "2563--2571",
• address = "Malaga, Spain",
• series = "GECCO '25 Companion",
• month = "14-18 " # jul,
• organisation = "SIGEVO",
• publisher = "Association for Computing Machinery",
• publisher_address = "New York, NY, USA",
• keywords = "genetic algorithms, genetic programming, ordinary differential equations, symbolic regression, system identification",
• isbn13 = "979-8-4007-1464-1",
• URL = "https://doi.org/10.1145/3712255.3734301",
• DOI = "doi:10.1145/3712255.3734301",
• size = "9 pages",
• abstract = "System identification is the task of automatically learning the model of a dynamical system, which can be represented as a system of ordinary differential equations (ODEs), using data points from time-series trajectories. This challenge has been addressed through various methods, including sparse regression, specialized neural networks, and symbolic regression. However, applying standard symbolic regression requires transforming the trajectory data to frame the problem as learning a set of regular equations. This study presents a first comprehensive comparison of the two most common data transformation approaches for system identification, evaluating their performance on a recently published benchmark suite of ODE systems. Our findings reveal that both approaches are highly sensitive to even moderate amounts of added noise, to different degrees. More surprisingly, we also show that data transformations can generate misleading search spaces, even under noise-free conditions. Further analysis indicates that reducing the data sampling step size significantly improves performance, suggesting that both transformation techniques are also affected by sampling frequency, and indicating possible future directions of research for system identification using symbolic regression.",
• notes = "GECCO-2025 SymReg workshop A Recombination of the 34th International Conference on Genetic Algorithms (ICGA) and the 30th Annual Genetic Programming Conference (GP)",

}

Genetic Programming entries for Alberto Tonda Hengzhe Zhang Qi Chen Bing Xue Mengjie Zhang Evelyne Lutton

Citations