Unsupervised Spiking Neural Network with Dynamic Learning of Inhibitory Neurons

. 2023 Aug 17;23(16):7232. doi: 10.3390/s23167232

Algorithm 1 Inhibition Weight Update and Synaptic Wiring

Require:
Weight from inhibition neuron in relation to excitatory layer $W_{S F}$ , synaptic wiring threshold $ϑ_{i n h i b i t}$ , inhibition weight learning rate $L_{1}$ and $L_{2}$ , maximum firing neuron $N_{m}$
1:
Set W_SF = − $50,$ $ϑ_{i n h i b i t}$ = − $38,$ L₁ $= 0.5,$ L₂ $= 0.07$
2:
// Inhibition weight update
3:
N_m = most active neuron among reacted excitatory neurons with $S n \geq 4$
4:
for ∀ reacted excitatory neurons based on the N_m and $W_{S F} < 0$ do
5:
if $S_{n} > S_{n - 1}$ then
6:
$W_{S F} \leftarrow W_{S F} + L_{1} (S_{n} - S_{n - 1})$
7:
else
8:
$W_{S F} \leftarrow W_{S F} + L_{2} (S_{n} - S_{n - 1})$
9:
end if
10:
// Synaptic wiring
11:
if $W_{S F} < 0$ and $W_{S F} > ϑ_{i n h i b i t}$ then
12:
$W_{S F} = + 25$
13:
end if
14:
end for