Here, I have the single hidden layer multi-layer perceptron (Input: 4 units, hidden unit: 3 units, output: 1 unit).

I want to do pruning before a training process with Keras. Concretely, I just want to remove the connection between the input units and hidden units. (so I don't remove the hidden units.)

Would you tell me how to implement this code? Now I am thinking that the skip connection with merge and concat is one of the solution, however, it is not smart way when the number of the units and layers are increased.

  • $\begingroup$ How can you have skip connections in such a shallow net? Plus it's more popular in CNN based archs as well. (You can add dropouts and checkout other variants on it as well) $\endgroup$ – Aditya Jun 3 '19 at 3:25

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.