Skip to main content
. 2020 Jun 30;14:653. doi: 10.3389/fnins.2020.00653

Figure 3.

Figure 3

(A) SNN computational graph unrolled over multiple time-steps for computing the V or Vmem and resultant output spike train S as a function of input spikes I[t] is shown. (B) Illustration of indeterminate derivative for a threshold or step function that is replaced with a surrogate piecewise linear derivative to allow the flow of gradients in AGD training.