search

nmi-lab / decolle-public

GNU General Public License v3.0
41 stars 22 forks source link

The DECOLLE loss seems not to be applied in the training process #22

Closed zjuls closed 2 years ago

zjuls commented 2 years ago

In the train_snn_bptt.py, the loss function is set as the CrossEntrophy function instead of the decolle loss in line 79. So how is the layer-wise decolle loss function applied during training?

eneftci commented 2 years ago

Note that cross entropy doesn't mean the losses are global, as cross entropy could also be used for local losses. But you are right that scripts under test use BPTT. This is because test is for testing the network in the best case training scenario, hence it uses BPTT. The decolle scripts (with local learning) are under scripts/.