Batch Normalization : What should we know?

First of all, we usually apply normalization to the data before throwing it into a model by making the variable subtracted with its mean and then divided by its standard deviation (Here, the normalization refers to the z score transformation). The use of normalization is to transform the numeric columns in the datasets to a more common scale without distorting the differences in the ranges of values or in other words without altering the skewness of the datasets. Well, sometimes, you may find out a more general way for calling normalization is feature scaling. Certainly, the main purpose of normalization is to reduce the time needed for the optimization of the parameters in a model. It is found out that the xy plane without normalization takes longer time to compute its optimum point because the variable with higher scale drags the progression.