I'm trying an experiment where I first train my model on MNIST labels [0-4], and then I freeze the first conv layers and continue training my model on labels [5-9]. When I dont re-initialize my Fully connected layer, the network seems to be stuck at 0 accuracy after 1 first epoch and then starts to learn efficiently.

What is the reason for this, why is the network stuck at 0 after 1 whole epoch ? on the other hand when I re-initialize the network it starts learning efficiently from the first epoch.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.