- author = "Krzysztof Krawiec and Tomasz Pawlak",
- title = "Approximating geometric crossover by semantic backpropagation",
- booktitle = "GECCO '13: Proceeding of the fifteenth annual conference on Genetic and evolutionary computation conference",
- year = "2013",
- editor = "Christian Blum and Enrique Alba and Anne Auger and Jaume Bacardit and Josh Bongard and Juergen Branke and Nicolas Bredeche and Dimo Brockhoff and Francisco Chicano and Alan Dorin and Rene Doursat and Aniko Ekart and Tobias Friedrich and Mario Giacobini and Mark Harman and Hitoshi Iba and Christian Igel and Thomas Jansen and Tim Kovacs and Taras Kowaliw and Manuel Lopez-Ibanez and Jose A. Lozano and Gabriel Luque and John McCall and Alberto Moraglio and Alison Motsinger-Reif and Frank Neumann and Gabriela Ochoa and Gustavo Olague and Yew-Soon Ong and Michael E. Palmer and Gisele Lobo Pappa and Konstantinos E. Parsopoulos and Thomas Schmickl and Stephen L. Smith and Christine Solnon and Thomas Stuetzle and El-Ghazali Talbi and Daniel Tauritz and Leonardo Vanneschi",
- isbn13 = "978-1-4503-1963-8",
- pages = "941--948",
- keywords = "genetic algorithms, genetic programming",
- month = "6-10 " # jul,
- organisation = "SIGEVO",
- address = "Amsterdam, The Netherlands",
-
DOI = "
doi:10.1145/2463372.2463483",
- publisher = "ACM",
- publisher_address = "New York, NY, USA",
- abstract = "We propose a novel crossover operator for tree-based genetic programming, that produces approximately geometric offspring. We empirically analyse certain aspects of geometry of crossover operators and verify performance of the new operator on both, training and test fitness cases coming from set of symbolic regression benchmarks. The operator shows superior performance and higher probability of producing geometric offspring than tree-swapping crossover and other semantic-aware control methods.",
-
notes = "GSGP.
'That is as geometric as possible' p942.
p943 targets ideal mixture of parents (rather than targeting the fitness function)
AGX Approximate matching (KD-tree, hashed) to exponentially large library of all possible (small) subtrees Exploit reversibility of common GP functions. Keeps multiple possible reverses but bounds are placed on potential exponential growth with crossover point distance from root node (ie depth).
p944 'no crossover at depth 17 was geometric'
Mutation required. Different types of selection.
Also known as \cite{2463483} GECCO-2013 A joint meeting of the twenty second international conference on genetic algorithms (ICGA-2013) and the eighteenth annual genetic programming conference (GP-2013)",
