- author = "Dirk Schweim and Franz Rothlauf",
- title = "An analysis of the bias of variation operators of estimation of distribution programming",
- booktitle = "GECCO '18: Proceedings of the Genetic and Evolutionary Computation Conference",
- year = "2018",
- editor = "Hernan Aguirre and Keiki Takadama and Hisashi Handa and Arnaud Liefooghe and Tomohiro Yoshikawa and Andrew M. Sutton and Satoshi Ono and Francisco Chicano and Shinichi Shirakawa and Zdenek Vasicek and Roderich Gross and Andries Engelbrecht and Emma Hart and Sebastian Risi and Ekart Aniko and Julian Togelius and Sebastien Verel and Christian Blum and Will Browne and Yusuke Nojima and Tea Tusar and Qingfu Zhang and Nikolaus Hansen and Jose Antonio Lozano and Dirk Thierens and Tian-Li Yu and Juergen Branke and Yaochu Jin and Sara Silva and Hitoshi Iba and Anna I Esparcia-Alcazar and Thomas Bartz-Beielstein and Federica Sarro and Giuliano Antoniol and Anne Auger and Per Kristian Lehre",
- pages = "1191--1198",
- address = "Kyoto, Japan",
-
DOI = "
doi:10.1145/3205455.3205582",
- publisher = "ACM",
- publisher_address = "New York, NY, USA",
- month = "15-19 " # jul,
- organisation = "SIGEVO",
- keywords = "genetic algorithms, genetic programming, EDA",
- isbn13 = "978-1-4503-5618-3",
-
abstract = "Estimation of distribution programming (EDP) replaces
standard GP variation operators with sampling from a
learned probability model. To ensure a minimum amount
of variation in a population, EDP adds random noise to
the probabilities of random variables. This paper
studies the bias of EDPs variation operator by
performing random walks. The results indicate that the
complexity of the EDP model is high since the model is
overfitting the parent solutions when no additional
noise is being used. Adding only a low amount of noise
leads to a strong bias towards small trees. The bias
gets stronger with an increased amount of noise. Our
findings do not support the hypothesis that sampling
drift is the reason for the loss of
diversity.
Furthermore, we suggest using property vectors to study the bias of variation operators. Property vectors can represent the distribution of a population's relevant property, such as tree depth or tree size. The Bhattacharyya coefficient of two property vectors is a measure of the similarity of the two distributions of population properties. The results for EDP and standard GP illustrate that search bias can be assessed by representing distributions using property vectors and measuring their similarity using the Bhattacharyya coefficient.",
- notes = "Also known as \cite{3205582} GECCO-2018 A Recombination of the 27th International Conference on Genetic Algorithms (ICGA-2018) and the 23rd Annual Genetic Programming Conference (GP-2018)",
