-4
$\begingroup$

Consider an example in which one wants process a 1d list of number with a CNN network, with 1d convolution.

Are the weight shared across the number of filters, or are they shared across the number of input feature maps ?

And are the weight used to compute the different feature maps, are they different or the same?

$\endgroup$

1 Answer 1

1
$\begingroup$

I'm not entirely clear what you're asking, but, let's say your convolutional input is $m$ channels of $n$ numbers. Assuming a same convolution. Then you output $d$ (maybe not the same as $m$) channels of $n$ numbers.

Do each of the $d$ output channels come from different filters with different weights? yes of course or they would be redundant. Are the same weights used for each of the $m$ input channels in each filter? no, not shared, so it can learn from each channel separately.

Is the convolutional filter the same across the $n$ numbers, when mapping input to output features? yes. Same filter weights are used across the input.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.