1Hardgeneral⏱ 45sWhy is batch normalization effective in training deep networks?📎▼AIt only helps with data loading speedBIt normalizes layer inputs to have stable mean/variance during training, reducing internal covariate shift and allowing higher learning rates with more stable, faster convergenceCIt replaces the need for any activation functionDIt only works for convolutional architectures