A spiking neural network with negative thresholding and its application to associative memory

This paper proposes a feedback-type spiking neural network using integrate-and-fire neurons with negative thresholding. Spiking neuron models express analog information by the timing of neuronal spike firing. Since these models operate asynchronously, it is expected that the spiking network operates faster than the conventional synchronous models. In order to apply such models to feedback networks with continuous states, spike firing expressing the zero value is required. However, the integrate-and-fire neurons generate no spikes when their internal potentials corresponding to the zero value do not exceed the threshold. To solve this problem, we propose negative thresholding that operates immediately after the spike input. In this paper, this thresholding operation is achieved by introducing a global excitatory unit. We have designed a spiking feedback network circuit with the global excitatory unit. The simulation results of associative memory using the network have revealed that the output spike-timing difference is independent of the input spike-timing range.