Simulation of Spiking Neural Nets with INSpiRE ME

While the usage of rate based neural nets for classification tasks is state of the art, this approach lacks one characteristic: time. However, many application patterns to be recognized are distributed over time. Although there exist some approaches introducing time into neural nets like the time delayed neural networks (TDNN), these approaches generally compute in cycles and are therefore insensitive to more subtle timescales. Thus, a more powerful approach is desirable. We examined spiking neural nets to tackle this task and developed an event-driven simulator for spiking neural nets called INSpiRE ME. The simulator uses the biologically inspired spike response model. To achieve a fast and powerful tool many design decisions had to be made, which will be explained in this paper.