What is gradient noise scale?

Updated May 16, 2026

Short answer

Gradient noise scale measures variability of stochastic gradients.

Deep explanation

It quantifies how noisy gradient estimates are compared to full-batch gradients. It helps determine optimal batch size and learning rate for efficient training.

Real-world example

Choosing batch size in training deep neural networks.

Common mistakes

  • Ignoring noise scale when scaling training systems.

Follow-up questions

  • Why is it important?
  • What affects noise scale?

More Gradient Descent interview questions

View all →