Batch Normalization
Normalizes layer inputs using batch statistics to stabilize and accelerate training.
In italiano: Batch NormalizationNormalizes using batch mean/variance, then scales/shifts with learnable parameters. Reduces internal covariate shift. Allows higher learning rates.
Examples
- After conv layers
- Before activation
- Stabilizing deep networks