What is batch normalization in PyTorch?

Updated May 17, 2026

Short answer

BatchNorm normalizes activations to stabilize and accelerate training.

Deep explanation

It normalizes layer inputs using batch mean and variance, then applies learnable scaling and shifting.

Real-world example

Used in CNNs to improve convergence speed.

Common mistakes

  • Using small batch sizes causing unstable statistics.

Follow-up questions

  • BatchNorm vs LayerNorm?
  • Why use running mean?

More PyTorch interview questions

View all →