We propose SFIMX, a method that reduces the number of required interactions between programs and tests in genetic programming. SFIMX performs factorization of the matrix of the outcomes of interactions between the programs in a working population and the tests. Crucially, that factorization is applied to matrix that is only partially filled with interaction outcomes, i.e., sparse. The reconstructed approximate interaction matrix is then used to calculate the fitness of programs. In empirical comparison to several reference methods in categorical domains, SFIMX attains higher success rate of synthesizing correct programs within a given computational budget.
@INPROCEEDINGS { Liskowski:2016:EuroGP,
ABSTRACT = { We propose SFIMX, a method that reduces the number of required interactions between programs and tests in genetic programming. SFIMX performs factorization of the matrix of the outcomes of interactions between the programs in a working population and the tests. Crucially, that factorization is applied to matrix that is only partially filled with interaction outcomes, i.e., sparse. The reconstructed approximate interaction matrix is then used to calculate the fitness of programs. In empirical comparison to several reference methods in categorical domains, SFIMX attains higher success rate of synthesizing correct programs within a given computational budget. },
ADDRESS = { Porto, Portugal },
AUTHOR = { Pawel Liskowski and Krzysztof Krawiec },
BOOKTITLE = { EuroGP 2016: Proceedings of the 19th European Conference on Genetic Programming },
EDITOR = { Malcolm I. Heywood and James McDermott and Mauro Castelli and Ernesto Costa },
KEYWORDS = { genetic algorithms, genetic programming, test-based problem, recommender systems, machine learning, surrogate fitness },
MONTH = { {30 } # mar # {--1 } # apr },
NOTES = { Part of \cite{Heywood:2016:GP} EuroGP'2016 held in conjunction with EvoCOP2016, EvoMusArt2016 and EvoApplications2016 },
ORGANISATION = { EvoStar },
PAGES = { 65--79 },
PUBLISHER = { Springer Verlag },
SERIES = { LNCS },
TITLE = { Surrogate Fitness via Factorization of Interaction Matrix },
VOLUME = { 9594 },
YEAR = { 2016 },
}