Hey Sagar,

You can implement the Flatten Layer pretty easily, you just need to follow the same steps as above.

The flatten layer does not have any trainable parameters, it’s only reshaping the input data. Therefore, only ∂E/∂X is required. And since no changes are made to the input, it’s just the ∂E/∂Y back to the original shape. You can check that it is mathematically correct.

MaxPool Layer needs a bit more calculus, but it’s feasible. I’ll let you try, hit me up if you struggle !