The authors present a neural network model that is able to learn and process sequential data. The strength of the model is that it renders its neurons with information processing capabilities and degrades its interconnections to simple one-bit delay lines. It involves neurons which exhibit spatio-temporal properties, pulse-coded delay lines as interconnections, and a local learning rule. The authors present the architecture and dynamics of a neuron, and the learning rule that was used to train simple networks. The model was successfully applied to learning the XOR function and to first-order conditioning.
展开▼
