Possible typo in the paper and some other questions

[zhujiagang]

An excellent work! I've several questions.
1, Is the equation (22) in paper https://arxiv.org/abs/1903.10520 right? I think the dot on H should be placed in the left of equation.
2, Are the equation (25) and (36) different? I think the square of l2norm of W{hat} (equation 36) is input channels instead of 1.
3, Can the WS be applied to fc layer, for example, the final linear layer of a classification model.
4, In WS, I think we should not focus on the gradients of W, because they doesn't affect the forward pass and backpropagation directly. Instead, we should focus on the gradients on X, which are directly affected by W{hat}. This might be the reason that WS works.
Sorry to bother you.