Skip to content
Technical GlossaryDeep Learning

Batch Normalization

A technique that normalizes intermediate activations at the mini-batch level to accelerate training and provide partial regularization.

Batch normalization helps reduce distribution shift inside deep networks and makes optimization more stable. By normalizing activations at the mini-batch level, it can enable training with higher learning rates. It may also introduce a mild regularization effect. It has played an important role in many architectures, from CNNs to pre-Transformer models.