Initializing weight with 0 for NN?

Initializing weights with 0s leads to neurons to learn the same features because the derivatives will remain same for every weight. Any same constant initialization will produce a poor result.