Replies: 1 comment 1 reply
-
|
The inputs don't have to be spikes. The idea is that the network will learn the optimal representation to convert non-spiking data into a format compatible with spiking. Tutorial 2 shows how non-spiking data is injected as a time-varying input current. The same logic applies. |
Beta Was this translation helpful? Give feedback.
-
|
Hi, while I get the idea, I don't understand - if the network learns the input spike representation, why do we still need time steps inside the forward function? Can it not be a single step and the network eventually learns generating correct spikes as it trains? I was also wondering where is the time axis in the input MNIST frames and why don't we convert them to spikes before passing to the network? Is there a problem? |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Description
train_loader = DataLoader(mnist_train, batch_size=batch_size, shuffle=True, drop_last=True)
test_loader = DataLoader(mnist_test, batch_size=batch_size, shuffle=True, drop_last=True)
It is using just this as input and there are no spikes
What I Did
Beta Was this translation helpful? Give feedback.
All reactions