Memetic Evolutionary Learning for Local Unit Networks

0345155 - ÚI 2011 RIV DE eng C - Konferenční příspěvek (zahraniční konf.)
Neruda, Roman - Vidnerová, Petra
Memetic Evolutionary Learning for Local Unit Networks.
Advances in Neural Networks – ISNN 2010. Vol. 1. Berlin: Springer, 2010 - (Zhang, L.; Lu, B.; Kwok, J.), s. 534-541. Lecture Notes in Computer Science, 6063. ISBN 978-3-642-13277-3. ISSN 0302-9743.
[ISNN 2010. International Symposium on Neural Networks /7./. Shanghai (CN), 06.06.2010-09.06.2010]
Grant CEP: GA ČR GA201/08/1744
Výzkumný záměr: CEZ:AV0Z10300504
Klíčová slova: radial basis function networks * evolutionary algorithms * memetic algorithms
Kód oboru RIV: IN - Informatika

In this work we propose two hybrid algorithms combining evolutionary search with optimization algorithms. One algorithm memetically combines global evolution with gradient descent local search, while the other is a two-step procedure combining linear optimization with evolutionary search. It is shown that these algorithms typically produce smaller local unit networks with performance similar to theoretically sound but large regularization networks.
Trvalý link: http://hdl.handle.net/11104/0186486