Correlation NN or Convolution NN

1

Hi,

I learned that Conv2d is actually a cross-correlation operator, hence, does that means our network is actually a Correlation NN instead of Convolution NN?

I can understand the difference between Correlation and Convolution is just a rotation of 180. However, what does it actually means when it comes to CNN? What if I change from convolution with correlation?

2

Well, we just call it “Convolution NN”.

According to my understanding(which could be wrong), whether you rotate the weights or not doesn’t matter(assume that after the rotation, it is still possible to use matrix multiplication) because to the NN, it is just weight.