1. 什么是Feature Normalization
standard deviation:标准差
variances: 方差
取出同一纬度i的所有x,计算它们的平均值记为
根据公式求得 放入这一维度中,这样他们的平均值为0,方差都是1,方便做gradient descent.
2. Considering Deep Learning
3.Batch Normalization- Testing
We do not always have batch at testing stage
Computeing the moving average of and
of the batches during traning
再拿上面的更新