0
$\begingroup$

I hope you are all well. I have to present an architecture of 1D CNN today and I am a bit confused. enter image description hereI have a 1D Convolutional neural network Consisting of input data, 3 fully connected 1D convolution layers, flatten layer, batch normalization layer, 2 dense layers, and one output layer with a softmax activation function. I am interested in the terminology of the layers because I have to represent them in a good way.I am interested in the terminology of how the layers are called. For example the 3 fully connected convolution layers according to my opinion can be represented as just Convolution Layers, the flattening and the batch normalization layers can be represented as post-processing layers, the two dense layers can be represented as Tickening Layers and the output layer with the softmax activation function can be represented as Activation Layer. This is how I have them at the moment, If you think that I am wrong somewhere please make a comment.

$\endgroup$
1
  • 1
    $\begingroup$ The activation layer is definitely misleading since CNNs and Dense Layers have also activation functions... $\endgroup$
    – malocho
    Jul 15 at 10:07
0
$\begingroup$

CNNs -> feature extraction

Flattering/Normalization -> downsampling

Denses -> fully connections

Softmax is actually an activation function, your CNNs have also activation functions. The softmax belongs to your dense layer part. When then I would call it as output layer

$\endgroup$
5
  • $\begingroup$ Hello sir, thanks for answering my question. I want to ask if you can provide me some information why the CNNs layers stands more for feature extraction rather than just convolution layers, flattering/normalization for downsampling and dense as fully connections layers. Thanks $\endgroup$
    – Ivan
    Jul 15 at 10:28
  • $\begingroup$ convolution layers learn different filters/kernels that extract features from your images $\endgroup$
    – malocho
    Jul 15 at 10:39
  • $\begingroup$ I am doing the CNN on time-series tabular data, not on an image dataset. $\endgroup$
    – Ivan
    Jul 15 at 10:40
  • $\begingroup$ it does not matter which type of data it is... it is about the construction of a convolution (image data was only an example) $\endgroup$
    – malocho
    Jul 15 at 10:43
  • $\begingroup$ Gaussian filter (as an example) can be applied on images as well as on time series... on both, it does smoothing... Imagine that convolution layers learn such different filters/kernels $\endgroup$
    – malocho
    Jul 15 at 10:45

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.