What is dropout and how does it work?

Updated May 17, 2026

Short answer

Dropout randomly disables neurons during training to prevent overfitting.

Deep explanation

It forces redundancy in representation by dropping activations with probability p during training.

Real-world example

Used in fully connected layers of deep networks.

Common mistakes

  • Using dropout during evaluation mode.

Follow-up questions

  • What happens in eval mode?
  • Why does dropout improve generalization?

More PyTorch interview questions

View all →