search

torch / nn

Other
1.34k stars 967 forks source link

Backprop procedure for nn.Narrow #1322

Closed pskrunner14 closed 5 years ago

pskrunner14 commented 5 years ago

Hi could anyone explain the backprop procedure for nn.Narrow module? As I understand, it is simply copying the gradients from the later layers into the indices of the input the operation was performed on during forward pass. Please correct me if I'm wrong. Thanks!