Improving Generalisation of AutoML Systems with Dynamic Fitness Evaluations

Created by W.Langdon from gp-bibliography.bib Revision:1.7615

@InProceedings{Evans:2020:GECCO,

• author = "Benjamin P. Evans and Bing Xue and Mengjie Zhang",
• title = "Improving Generalisation of {AutoML} Systems with Dynamic Fitness Evaluations",
• year = "2020",
• editor = "Carlos Artemio {Coello Coello} and Arturo Hernandez Aguirre and Josu Ceberio Uribe and Mario Garza Fabre and Gregorio {Toscano Pulido} and Katya Rodriguez-Vazquez and Elizabeth Wanner and Nadarajen Veerapen and Efren Mezura Montes and Richard Allmendinger and Hugo Terashima Marin and Markus Wagner and Thomas Bartz-Beielstein and Bogdan Filipic and Heike Trautmann and Ke Tang and John Koza and Erik Goodman and William B. Langdon and Miguel Nicolau and Christine Zarges and Vanessa Volz and Tea Tusar and Boris Naujoks and Peter A. N. Bosman and Darrell Whitley and Christine Solnon and Marde Helbig and Stephane Doncieux and Dennis G. Wilson and Francisco {Fernandez de Vega} and Luis Paquete and Francisco Chicano and Bing Xue and Jaume Bacardit and Sanaz Mostaghim and Jonathan Fieldsend and Oliver Schuetze and Dirk Arnold and Gabriela Ochoa and Carlos Segura and Carlos Cotta and Michael Emmerich and Mengjie Zhang and Robin Purshouse and Tapabrata Ray and Justyna Petke and Fuyuki Ishikawa and Johannes Lengler and Frank Neumann",
• isbn13 = "9781450371285",
• publisher = "Association for Computing Machinery",
• publisher_address = "New York, NY, USA",
• URL = "https://doi.org/10.1145/3377930.3389805",
• DOI = "doi:10.1145/3377930.3389805",
• booktitle = "Proceedings of the 2020 Genetic and Evolutionary Computation Conference",
• pages = "324--332",
• size = "9 pages",
• keywords = "genetic algorithms, genetic programming, regularized evolution, AutoML, regularization, automated machine learning, generalisation, dynamic fitness evaluations",
• address = "internet",
• series = "GECCO '20",
• month = jul # " 8-12",
• organisation = "SIGEVO",
• abstract = "A common problem machine learning developers are faced with is overfitting, that is, fitting a pipeline too closely to the training data that the performance degrades for unseen data. Automated machine learning aims to free (or at least ease) the developer from the burden of pipeline creation, but this overfitting problem can persist. In fact, this can become more of a problem as we look to iteratively optimise the performance of an internal cross-validation (most often k-fold). While this internal cross-validation hopes to reduce this overfitting, we show we can still risk overfitting to the particular folds used. In this work, we aim to remedy this problem by introducing dynamic fitness evaluations which approximate repeated k-fold cross-validation, at little extra cost over single k-fold, and far lower cost than typical repeated k-fold. The results show that when time equated, the proposed fitness function results in significant improvement over the current state-of-the-art baseline method which uses an internal single k-fold. Furthermore, the proposed extension is very simple to implement on top of existing evolutionary computation methods, and can provide essentially a free boost in generalisation/testing performance.",
• notes = "Also known as \cite{10.1145/3377930.3389805} GECCO-2020 A Recombination of the 29th International Conference on Genetic Algorithms (ICGA) and the 25th Annual Genetic Programming Conference (GP)",

}

Genetic Programming entries for Benjamin Evans Bing Xue Mengjie Zhang

Citations