30.11.2007: HIIT seminar: Jarkko Tikka

HIIT seminars in fall 2007 will be held in hall **B222** of Exactum,
on Fridays starting at 10:15 a.m. Coffee available from 10.

Fri Nov 30
Jarkko Tikka
Two approaches to select relevant inputs for RBF network


In regression problems, making accurate predictions is often the primary goal. Neural networks provide good generalization in many cases, but their interpretability is usually limited. The dependencies between input
and output variables is not clear at all. However, the contributions of input variables in the prediction of output would be valuable information in many real world applications. Two input selection algorithms for Radial Basis Function (RBF) networks are presented to increase the interpretability of the models. Firstly, a sequential input selection algorithm is proposed. It is a backward selection type algorithm based on the partial derivatives of the RBF network. Secondly, a input selection method, which is based on a constrained cost function, is shown. The input selection algorithms are applied to both simulated and benchmark data sets and the results are convincing. That is, both algorithms yield accurate prediction models, which are also parsimonious in terms of input variables.


Last updated on 27 Nov 2007 by Teemu Mäntylä - Page created on 30 Nov 2007 by Teija Kujala