%0 Journal Article %1 Zhang:1998:eisnt %A Zhang, Byoung-Tak %A Ohm, Peter %A Mühlenbein, Heinz %D 1997 %J Evolutionary Computation %K Minimum algorithm algorithms, breeder description genetic higher-order induction, length networks, neural prediction, principle, program programming, representation, series time tree %N 2 %P 213--236 %R doi:10.1162/evco.1997.5.2.213 %T Evolutionary Induction of Sparse Neural Trees %U http://www.mitpressjournals.org/doi/pdfplus/10.1162/evco.1997.5.2.213 %V 5 %X This paper is concerned with the automatic induction of parsimonious neural networks. In contrast to other program induction situations, network induction entails parametric learning as well as structural adaptation. We present a novel representation scheme called neural trees that allows efficient learning of both network architectures and parameters by genetic search. A hybrid evolutionary method is developed for neural tree induction that combines genetic programming and the breeder genetic algorithm under the unified framework of the minimum description length principle. The method is successfully applied to the induction of higher order neural trees while still keeping the resulting structures sparse to ensure good generalization performance. Empirical results are provided on two chaotic time series prediction problems of practical interest.

@article{Zhang:1998:eisnt, abstract = {This paper is concerned with the automatic induction of parsimonious neural networks. In contrast to other program induction situations, network induction entails parametric learning as well as structural adaptation. We present a novel representation scheme called neural trees that allows efficient learning of both network architectures and parameters by genetic search. A hybrid evolutionary method is developed for neural tree induction that combines genetic programming and the breeder genetic algorithm under the unified framework of the minimum description length principle. The method is successfully applied to the induction of higher order neural trees while still keeping the resulting structures sparse to ensure good generalization performance. Empirical results are provided on two chaotic time series prediction problems of practical interest.}, added-at = {2008-06-19T17:35:00.000+0200}, author = {Zhang, Byoung-Tak and Ohm, Peter and M{\"u}hlenbein, Heinz}, biburl = {https://www.bibsonomy.org/bibtex/20d10fbb72a3cfacf8b038ab78c4be2f3/brazovayeye}, doi = {doi:10.1162/evco.1997.5.2.213}, interhash = {314b58714d323e98d1ff4c0796b37a4d}, intrahash = {0d10fbb72a3cfacf8b038ab78c4be2f3}, journal = {Evolutionary Computation}, keywords = {Minimum algorithm algorithms, breeder description genetic higher-order induction, length networks, neural prediction, principle, program programming, representation, series time tree}, notes = {Evolutionary Computation (Journal) Special Issue: Trends in Evolutionary Methods for Program Induction Referenced in \cite{zhang:1997:WSC2} Demonstrated on Mackey-Glass and chaotic fluctuations in a far-infared ammonia NH3 laser. Libraries of building blocks (selected by their local fitness), local fitness-based crossover, injection and pruning of submodules (subtree replaced by its descendent subtree, if the descendent is fitter than the subtree itself), scheduling of genetic operators. Parsimony fitness bias. Local search to optimise sigma+pi? neural net weights using exponetial noise.}, number = 2, pages = {213--236}, size = {31 pages}, timestamp = {2008-06-19T17:55:17.000+0200}, title = {Evolutionary Induction of Sparse Neural Trees}, url = {http://www.mitpressjournals.org/doi/pdfplus/10.1162/evco.1997.5.2.213}, volume = 5, year = 1997 }