I saw some implementations of Yoon Kim's Convolutional Neural Network (Paper: http://www.aclweb.org/anthology/D14-1181)....

...in some implementations they put one more Dense(..) Layer before the output layer (with softmax activation). In my opinion there is no additional Dense(..) Layer in the original paper:

So what's the original implementation?

differing Yoon Kim CNN implementations


Based on the model architecture from original paper below, it looks like there is just one dense layer at the end. Conv layers --> Max Pool over time --> Dense

enter image description here

| improve this answer | |
  • 1
    $\begingroup$ I‘m of the same opinion. So would you put the dropout after the flatten()? $\endgroup$ – Predicted Life Oct 8 '19 at 15:52
  • $\begingroup$ Yes, I think dropout comes after flatten. I implemented this in Pytorch, and I used Conv --> Maxpool --> Concat (in this case it is similar to flatten in Keras) --> Dropout --> Dense $\endgroup$ – Tamirlan Oct 8 '19 at 16:34

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.