Novel Constructions are presented which allow the spike-based computation of various arithmetic functions. These constructions, which are binary I/O stochastic automata with an internal counter state, are able to give faithful reproductions of the desired function whenever the output would fall into the range (0, 1), and automatic clipping to 0 or 1 otherwise. In is demonstrated how such arithmetic modules can be used to provide an output transform function for pulse-processing digital neural networks; many other applications within the field of sotchastic neural computing are possible.
展开▼
