Temporal codes for pulsed networks

Pulsed networks provide a novel set of architectures for the implementation and design of neural network computing paradigms. The author considers pulsed networks in which the pulses do not vary in amplitude or duration, but convey information solely by the placement of impulses in time. The author addresses how information can be represented and encoded in such signals, and presents alternative codes for neural net pulse train signals. Pulsed codes include firing rate, interval/pattern codes, and ensemble synchrony codes. In firing rate codes, different firing rates represent the information to be carried. Although firing rates are the simplest scheme, they take more time to be received and have low amounts of accuracy for distinctions between different levels of firing rate. Interval/pattern codes, consisting of two or more pulses spaced by particular timing intervals, allow for large numbers of coded words. Especially when patterns are comprised of segments of multiunit impulse trains, but require special post-processing circuits. Ensemble synchrony codes, in which multiple units fire in near-synchrony, can allow n units to carry more than 2^n/2 signals, and do not require special circuits to detect the synchronies