Your code is pretty good. But I have a question. The paper mentions that the convolutional layers are inspired by Winner-Take-All autoencoders. I tried by myself to add WTA autoencoder to the last convolutional layer but failed. The training cost was very high after applying WTA autoencoder . Do you have any idea about this problem?
Hi,
Your code is pretty good. But I have a question. The paper mentions that the convolutional layers are inspired by Winner-Take-All autoencoders. I tried by myself to add WTA autoencoder to the last convolutional layer but failed. The training cost was very high after applying WTA autoencoder . Do you have any idea about this problem?