Does input need to be standardized if you use batch normalization?

Question

I play with the normalization of the party in Keras. I was wondering if normalization of dosage normalizes to the neural network. Does this mean that I do not need to standardize my network logins and rely on BN for this?

Answer 1

Although you can use it for this, it is not intended for this, and you are likely to introduce a sampling error in your normalization due to the limited sample size (sample size is the size of your batch).

Another factor, why I would not recommend using the standard batch for this, is that it introduces the corrective terms gamma and beta (prepared parameters) that will distort your training data if they are not disabled.

To normalize your test data, I would recommend using the z-score normalization in the full set of training sessions or something suitable, but not normalizing the game.