What is an activation function?

Updated May 17, 2026

Short answer

An activation function introduces non-linearity into a neural network.

Deep explanation

Without activation functions, the network becomes a linear regression model regardless of depth. Common activations include ReLU, sigmoid, and tanh.

Real-world example

ReLU is used in CNNs for image classification.

Common mistakes

  • Using sigmoid in deep networks causing vanishing gradients.

Follow-up questions

  • Why is ReLU preferred?
  • What is dying ReLU?

More Neural Networks interview questions

View all →