Batch Normalization

Technique that normalizes layer inputs during training, reducing internal covariate shift and improving training stability and speed.