A preliminary comparison of FIR-synapse neural networks with and without local output feedback
N. Kuehner, Maryhelen Stevenson · 2002
In the study of neural network architectures, a wide variety of neuron structures and interconnection schemes have been proposed and studied. The article looks at a multilayered globally feedforward network architecture in which each neuron has potential for local output feedback and where all interconnections, both feedforward and feedback, are made by FIR filters. Issues of stability and convergence, speed, efficiency, and generality are investigated. The applications of this architecture to problems of system identification are described in an attempt to assess its strengths and weaknesses in comparison to the standard feedforward FIR network.