A neural network for generating diverse decision boundaries with a minimal number of parameters

We present a neural network architecture that can provide a large number of flexible decision bundarie for pattern recognition using a minimal number of parameters. A biologically compatible implementation of the network leads to greater discrimination power.