Dear all, As I'm trying to build an RNN model based on the feature from CNN, I implemented the model in Theano. But I found that the training error stops to decrease at very early stage with very high error. I searched this kind of joint training problem, almost all of the implementations leverage the pre-trained CNN, rather than train CNN and RNN jointly from scratch. So, why it's difficult to directly train CNN and RNN jointly? Could any body give some deep thoughts about this problem?
Thanks, Xin -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
