Krzysztof Krawiec


Home

Research:

edit SideBar

We propose an alternative program representation that relies on automatic design of semantic-based embeddings of genetic programs into discrete multidimensional spaces. The embedding imposes a well-structured hypercube topology on the search space, endows it with a semantic-aware neighbourhood, and enables convenient search using Cartesian coordinates. The embedding algorithm consists in locality-driven optimization and operates in abstraction from a specific fitness function, improving locality of all possible fitness landscapes simultaneously. In experimental part, we validate the approach on the domain of symbolic regression and demonstrate on a large sample of problem instances that semantic embedding provides better search performance than the original program space. Moreover, we show also that semantic embedding of small programs can be effectively exploited in a compositional manner to search the space of large compound programs. Krawiec11Bib

@INPROCEEDINGS { DBLP:conf/gecco/Krawiec11,
    AUTHOR = { Krzysztof Krawiec },
    TITLE = { Semantically embedded genetic programming: automated design of abstract program representations },
    ABSTRACT = { We propose an alternative program representation that relies on automatic design of semantic-based embeddings of genetic programs into discrete multidimensional spaces. The embedding imposes a well-structured hypercube topology on the search space, endows it with a semantic-aware neighbourhood, and enables convenient search using Cartesian coordinates. The embedding algorithm consists in locality-driven optimization and operates in abstraction from a specific fitness function, improving locality of all possible fitness landscapes simultaneously. In experimental part, we validate the approach on the domain of symbolic regression and demonstrate on a large sample of problem instances that semantic embedding provides better search performance than the original program space. Moreover, we show also that semantic embedding of small programs can be effectively exploited in a compositional manner to search the space of large compound programs. },
    BIBSOURCE = { DBLP, http://dblp.uni-trier.de },
    EE = { http://doi.acm.org/10.1145/2001576.2001762 },
    BOOKTITLE = { Proceedings of the 13th Annual Conference on Genetic and Evolutionary Computation },
    SERIES = { GECCO '11 },
    YEAR = { 2011 },
    ISBN = { 978-1-4503-0557-0 },
    LOCATION = { Dublin, Ireland },
    PAGES = { 1379--1386 },
    NUMPAGES = { 8 },
    URL = { http://doi.acm.org/10.1145/2001576.2001762 },
    DOI = { 10.1145/2001576.2001762 },
    ACMID = { 2001762 },
    PUBLISHER = { ACM },
    ADDRESS = { New York, NY, USA },
    KEYWORDS = { genetic programming, genotype-phenotype mapping, locality, program representation, program semantics },
}


Powered by PmWiki