Skip to main content
. 2019 Mar 19;13:189. doi: 10.3389/fnins.2019.00189

Figure 4.

Figure 4

MNIST digit representations (re-arranged in 28 × 28 format) learnt by the synapses connecting the input to each excitatory neuron in a binary fully-connected SNN of 400 neurons (arranged in 20 × 20 grid). The binary fully-connected SNN is trained using (A) the proposed eHB-STDP containing distinct Hebbian potentiation and anti-Hebbian depression windows separated by a dead zone, (B) eHB-STDP2 where the dead zone is replaced with a wider Hebbian potentiation window, and (C) eHB-STDP3 where the dead zone is replaced with a wider anti-Hebbian depression window.