Spike specialized dropout layers

[kimborgen]

The dropout layers cause the spikes (1) to scale to 1.42... Que? When I remove the dropout its ok spk_conv1 = self.dropout1(spk_conv1)

[kimborgen]

Ah, torch.dropout scales the spikes! https://pytorch.org/docs/stable/generated/torch.nn.Dropout.html

Because scaling is required for ANN but does not work for spikes, lets remove the dropout layers for now and research spike specific dropout methods.