Study on general second-order neural units (SONUs)

In this paper, a general second-order neural unit (SONU) is developed using a new matrix form which can provide a general second-order combination of the input signals and synaptic weights. It is shown that, from the point of view of both the neural computing process and its learning algorithm, the linear combination neural units used widely in multilayer neural networks are only a subset of the proposed SONUs. Simulation studies for both the pattern classification and function approximation problems demonstrate that the learning and generalization abilities of the proposed SONUs are much superior to that of the linear combination neural units.