A new neural network architecture based on quadratic function neurons

In this paper, a class of multilayer perceptrons known as rotational quadratic function neural networks (RQFNN) is introduced. The rotational quadratic function neuron (RQFN), at the center of this class of networks, is a particular implementation of the quadratic function neuron (QFN). Compared with the traditional implementation, the RQFN requires much less fan-ins and thus much smaller cross-connection volume. The economy of the fan-ins and the cross connection volumes facilitates the mapping of the model onto silicon. >