Input Space Division by Local PCA for Classification with Multilayer Perceptrons

E. López-Rubio, J.A. Gómez-Ruiz, and I. López-Rubio (Spain)


Neural networks, local Principal Components Analysis,multilayer perceptrons, input space division.


Multilayer perceptrons (MLPs) have been widely used in supervised classification. While it is well known that a MLP is a universal approximator, it is also well known that the approximation may be very greedy. Here we perform an input space division by a local Principal Components Analysis (PCA) network in order to improve the classification performance of MLPs. Experimental results are presented, which show that our hybrid approach outperforms a standard MLP classification architecture in some well-known benchmark problems.

Important Links:

Go Back