Algorithm 1.

Forward propagation with dropout at each iteration in SNN

1:  Input : Poisson-distributed input spike train (inputs), Dropout ratio (p), Total number of time steps (#timesteps), Membrane potential (Vmem), Time constant of membrane potential (τm), Firing      threshold (Vth)

2:  Initialize $SN{N}^l.V_{mem}\leftarrow 0Al=2,\dots ,\#SNN.layer$

3:  // Define the random subset of units (with a probability 1 − p) at each iteration

4:  for l ← 1 to #SNN.layer − 1 do

5:        maskl ← generate_random_subset(probability = 1 − p)

6:  for t ← 1 to #timesteps do

7:       // Set input of first layer equal to spike train of a mini-batch data

8:       SNN1.spike[t] ← inputs[t];

9:       for l ← 2 to #SNN.layer do

10:           // Integrate weighted sum of input spikes to membrane potential

11:           $SN{N}^l.V_{mem}\left[t\right]\leftarrow SN{N}^l.V_{mem}\left[t-1\right]+SN{N}^{l-1}forward(SN{N}^{l-1}.spike\left[t\right]).*(mas{k}^{l-1}/(1$-p));

12:           // If Vmem is greater than Vth, post-neuron generate a spike

13:           if $SN{N}^l.V_{mem}\left[t\right]>SN{N}^l.V_{th}$ then

14:                // Membrane potential resets if the corresponding neuron fires a spike

15:                SNNl.spike[t] ← 1

16:                $SN{N}^l.V_{mem}\left[t\right]\leftarrow 0$

17:           else

18:                // Else, membrane potential decays over time

19:                SNNl.spike[t] ← 0

20:                $SN{N}^l.V_{mem}\left[t\right]\leftarrow e^{-\frac{1}{{\tau }_m}}*SN{N}^l.V_{mem}\left[t\right]$