Experimental Evaluation of Generalization Capability of Incremental Projection Learning in Neural Networks

Hendri Murfi
hendri@makara.cso.ui.ac.id
Mathematics
University of Indonesia

Indonesia

Abstract

One of the essences of learning in neural network is generalization capability. It is an ability to give an accurate result for data that was not learned in learning process. One of learning method that has goal to increase generalization capability directly is incremental projection learning. This learning method use analysis functional approach in reproducing kernel Hilbert space. This paper will describe experimental evaluation of generalization capability of incremental projection learning. Then, Make comparison with other learning method, i.e. incremental back propagation learning and incremental radial basis learning. Base on our experiment, incremental projection learning gives better generalization capability when the number of training examples is small and the noise variance is large. The other case, it does not always give the better generalization capability. Even though, In case learning data are big enough and noise variance are small enough, incremental projection learning always gives worse generalization capability than incremental back propagation learning. Our experiment used Matlab. Key Words: generalization capability, Incremental projection learning, neural network


© ATCM, Inc. 2001.