This is not an issue or problem, i just want to share an interesting thing i found about convolutional model, that the backward pass through a convolution (or the derivative of J w.r.t A and W) are also 2 convolutions! More specific:
dW = convolution of A and dZ
dA = convolution of W (with padding) and 180degree rotation of dZ
I’m not an expert on algebra so it is very hard for me to explain the idea, luckily i found an article on Medium about it: How does Backpropagation work in a CNN? | Medium (thanks to the writer - Pavithra Solai)
Hope it gives you some intuition about convolution and the beauty behind it!