Handwritten digits recognition system via OCON neural network by pruning selective update

Abstract

Performs the handwritten digits recognition using the OCON (one-class-one-net) network and the PSU (pruning selective update) training algorithm. The main feature of the architecture of OCON network is that the entire network is composed of single output multi-layer perceptron and each of the subnets represents one class. The PSU training algorithm defined on the new cost function is designed to speed up the training procedure. It is shown that an OCON network with the new training algorithm outperforms the conventional back-propagation algorithm.<>