I am trying to reduce a conv output of x channels to y channels through a sum reduction. For example, if I have a output of (batches, h, w, d, 15), I want to reduce it to (batches, h, w, d, 5). I could not find a way to do this with the tf.math.reduce() function.

I tried using tf.math.reduce(), but it does not have a way to specify how many dimensions to keep. I also tried to reshape the tensor, but was not able to figure it out.