Batch normalization rescales the values flowing between layers of a neural network during training, which can allow higher learning rates and faster training.
In plain words
As a network trains, the numbers passed from one layer to the next keep shifting, which makes learning slower and fussier. Batch normalization adjusts those numbers for each small batch of training examples so they stay in a steady range. That lets training use larger steps and can reduce overfitting.
A quick example
Imagine an image model that trains slowly and becomes unstable when the learning rate is raised. Adding batch normalization keeps each layer's values in a steady range, so the team can try a higher learning rate and reach good accuracy in fewer steps.
Where you'll meet it
- Image and other deep neural networks during training.
- Architecture diagrams, often shortened to BatchNorm.
Related
Sources
- Docsbatch normalization, Google Machine Learning Glossary · read 29 Sept 2026
- PaperBatch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift, Ioffe and Szegedy (arXiv) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.