when computing dA the formula used is dA = sum(Wc * dZ)
I read a little about computing dA, from which I saw that dA = W_flipped * Z
which is the true/ mathematicaly convolution during backprop (and using padding)
why in the code nothing is mentioned about flipping and how does it work in first place with out flipping ?
The formulas as presented here are consistent. Prof Ng uses the convention that the gradient of a tensor has the same shape as the base tensor. In the “pure math” version of all this, they define it in a different way such that the gradient is transposed.
If you just follow Prof Ng’s way of doing things, it all works. Start by comparing the dimensions in the formulas and you’ll see it’s all consistent.