A pulse transmission neural network architecture that recognizes patterns and temporally integrates

Summary form only given, as follows. The author has developed a pulse transmission (PT) neural network architecture based on a model with a simplified biological synapse. Neurons transmit pulses, not numbers, to one another. Each neuron integrates incoming pulses over time and generates on outgoing pulse when its activity level reaches threshold. Weights are associated with each interconnection, and dictate the amount of influence on the target cell for each arriving impulse. Inputs and outputs of the network are spatiotemporal patterns of pulses. An XOR function has been calculated with this model, as well as pattern classification of coarse grid patterns. The network can temporally integrate arriving signals, so that an incoming pattern can arrive over a period of time and still be recognized. The network also displays some resilience to temporal noise-noise in which segments of an arriving pattern are shifted in time