0
$\begingroup$

Let's say that I have a convolutional neural network with multiple blocks, each consisting of multiple filters. If we have something along the lines of Input -> Block1 -> Block2 -> Block3 -> Max Pooling -> Softmax. Now, I deleted Block2. How would I connect Block1 architecturally to Block3?

If I were to delete Block3, what can I do?

Please let me know if it would be best if I gave these blocks numbers, each with a number of filters.

$\endgroup$
  • $\begingroup$ You can attach any part of the network to another part, but you will have to learn the new connections that connect them. An example of this is transfer learning. $\endgroup$ – Frans Rodenburg Sep 24 '19 at 9:48
0
$\begingroup$

Assuming you have appropriate padding and stride=1, and the number of filters in each layer are the same (since you don't say otherwise...) such that other than the max pool layer, the feature maps are all of the same sizes:

Simply pass the output of block 1 to the input of block 3. It's unclear what you mean by "connect" -- but I guess this counts as the identity connection. If you delete block 3 then simply pass the output of the previous block to the max pool layer (again using the "identity connection").

$\endgroup$

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.