- author = "R. Muhammad Atif Azad and David Medernach and Conor Ryan",
- title = "Efficient interleaved sampling of training data in genetic programming",
- booktitle = "GECCO Comp '14: Proceedings of the 2014 conference companion on Genetic and evolutionary computation companion",
- year = "2014",
- editor = "Christian Igel and Dirk V. Arnold and Christian Gagne and Elena Popovici and Anne Auger and Jaume Bacardit and Dimo Brockhoff and Stefano Cagnoni and Kalyanmoy Deb and Benjamin Doerr and James Foster and Tobias Glasmachers and Emma Hart and Malcolm I. Heywood and Hitoshi Iba and Christian Jacob and Thomas Jansen and Yaochu Jin and Marouane Kessentini and Joshua D. Knowles and William B. Langdon and Pedro Larranaga and Sean Luke and Gabriel Luque and John A. W. McCall and Marco A. {Montes de Oca} and Alison Motsinger-Reif and Yew Soon Ong and Michael Palmer and Konstantinos E. Parsopoulos and Guenther Raidl and Sebastian Risi and Guenther Ruhe and Tom Schaul and Thomas Schmickl and Bernhard Sendhoff and Kenneth O. Stanley and Thomas Stuetzle and Dirk Thierens and Julian Togelius and Carsten Witt and Christine Zarges",
- isbn13 = "978-1-4503-2881-4",
- keywords = "genetic algorithms, genetic programming: Poster",
- pages = "127--128",
- month = "12-16 " # jul,
- organisation = "SIGEVO",
- address = "Vancouver, BC, Canada",
-
URL = "
http://doi.acm.org/10.1145/2598394.2598480",
-
DOI = "
doi:10.1145/2598394.2598480",
- publisher = "ACM",
- publisher_address = "New York, NY, USA",
-
abstract = "The ability to generalise beyond the training set is
important for Genetic Programming (GP). Interleaved
Sampling is a recently proposed approach to improve
generalisation in GP. In this technique, GP alternates
between using the entire data set and only a single
data point. Initial results showed that the technique
not only produces solutions that generalise well, but
that it so happens at a reduced computational expense
as half the number of generations only evaluate a
single data point.
This paper further investigates the merit of interleaving the use of training set with two alternatives approaches. These are: the use of random search instead of a single data point, and simply minimising the tree size. Both of these alternatives are computationally even cheaper than the original setup as they simply do not invoke the fitness function half the time. We test the utility of these new methods on four, well cited, and high dimensional problems from the symbolic regression domain.
The results show that the new approaches continue to produce general solutions despite taking only half the fitness evaluations. Size minimisation also prevents bloat while producing competitive results on both training and test data sets. The tree sizes with size ionisation are substantially smaller than the rest of the setups, which further brings down the training costs.",
- notes = "Also known as \cite{2598480} Distributed at GECCO-2014.",
