Multi-Objectivization and Surrogate Modelling for Neural Network Hyper-Parameters Tuning

Autoři: Mgr. Roman Neruda, Ph.D.
Rok vydání: 2013
Typ publikace: Konferenční příspěvek (mezinárodní konf.)
Periodikum: Emerging Intelligent Computing Technology and Applications
Projekty: Modelování složitých systémů softcomputingovými metodami
Bibliografická citace: PILÁT, M., NERUDA, Roman. Multi-Objectivization and Surrogate Modelling for Neural Network Hyper-Parameters Tuning. In: HUANG, D.S., GUPTA, P., WANG, L., GROMIHA, M.M., eds. Emerging Intelligent Computing Technology and Applications. Berlin: Springer, 2013, S. 61-66. Communications in Computer and Information Science, 375. ISBN 978-3-642-39677-9. ISSN 1865-0929
Dostupné z: doi:10.1007/978-3-642-39678-6_11
DOI: 10.1007/978-3-642-39678-6_11
ID: 428789

Abstrakt

We present a multi-objectivization approach to the parameter tuning of RBF networks and multilayer perceptrons. The approach works by adding two new objectives - maximization of kappa statistic and minimization of root mean square error - to the originally single-objective problem of minimizing the classification error of the model. We show the performance of the multi-objectivization approach on five data sets and compare it to a surrogate based single-objective algorithm for the same problem. Moreover, we compare the multi-objectivization approach to two surrogate based approaches - a single-objective one and a multi-objective one.