0
$\begingroup$

In the field of text classification, it is common to use Conv1D filters running over word embeddings and then getting a single value on the output for each filter using GlobalMaxPooling1D.

As I understand the process, the convolutional filter is a matrix of the same size as the $$\text{size of filter matrix} = \text{embedding dim}\cdot\text{width of the filter}$$ The filter matrix is then applied to the input embeddings (multiplied element by element) which produces a matrix of the same size for each filter position. Not a single number.

So how does the global max pooling get a single number on the output? Does it simply take a maximum over all the values in all the output matrices, or is there any other processing?

Please correct me if I'm wrong.

$\endgroup$
0
$\begingroup$

Apparently I forgot how the convolution works. The input is multiplied element wise with the filter weights and the products are then summed. That's how a single value is obtained on the output.

$\endgroup$

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.