Batch Normlization

Make a fool of each feature individually , So that the mean value of each feature is 0, variance 1 Distribution
Although the whitening operation of each feature reduces the expression ability of input data at each layer of the network, no linear transformation operation is added
On the data min-batch On the basis of Batch Normalization

The formula ：
> For the first neuron $u_1$=1.5, $eth{a}_1^2$ = 0.44 Use the above formula to normalize the first row of data .

• For use in the test phase
  utilize BN Train the model , Keep each group min-batch Training data is on each layer $U_{t}est$ , $eth{a}_{t}est$, Use the statistics of the whole sample to Test The data were normalized Make unbiased estimation