A neural network structure and learning algorithms with the neuron output feedback

In this paper, a new neuron model with different output-feedback factor and a neural network model that is composed of output feedback neural model are proposed. And its learning algorithm is derived and proofed in theory. This neural network can learn not only the static knowledge, but can learn dynamic knowledge; not only can remember static information, but can remember dynamic information; so that it is a truly dynamic neural network. Based on the minimum variance theory, the learning algorithm of the neural network with output feedback is proofed in theory. And the algorithm is summed up in the form of the theorem. Theoretical studies have shown that the static weights indicate the performance of the static mapping. Neuron output feedback coefficient implies the dynamic evolution performance of the neural network. And different feedback coefficients express the dynamic performance of different neurons. Therefore, the Study for dynamic characteristics and learning strategies of output feedback neural network is of great theoretical significance and application value.