I'm studying the gradient descent algorithm for single hidden layer neural networks. Suppose that I have an initial dataset and then I use mean normalization in order to scale the features.

Why mathematically do the normalized features carry the same information of the initial features?