Abstract:
A two-layer neural network called an extended differentiator network (EDN), which combines unsupervised and supervised training, is presented The EDN uses a soft competitive learning method in the unsupervised layer followed by a supervised associative layer. The soft competitive learning in the EDN takes the activity of all the competing neurons into account by using a one-step lateral inhibition mechanism. The functionality of the network is tested on a vowel recognition task and a cluster analysis problem. The simulation results indicate an effective use of the competing neurons, resulting in a high recognition rate in a network with a simple configuration