In this seminar we present some recent results from experiments in the investigation of regulating the growth of a particular type of evolving connectionist system, namely the Evolving Fuzzy Neural Network (EFuNN). This work in part addresses previous research which speculated that this evolving connectionist system could be further developed with a view to either reducing the overall number of learning parameters or having them adjusted automatically in order to regulate the growth of the EFuNN structure.
One of these areas is in the pruning of the EFuNN and in this presentation we discuss the originally proposed pruning method and offer an alternative method in which we apply an entropy criterion to automatically regulate its growth. We test this alternative method against two benchmark classification data sets and the results of the experiments suggest that this new method performs better than the originally proposed method of pruning an EFuNN. We also make some preliminary comments on its efficacy with time series data sets.
Last modified: Monday, 11-Jul-2011 14:21:31 NZST
This page is maintained by the seminar list administrator.