What is sparse autoencoder and why is it used?

Sparse autoencoders enforce sparsity in hidden activations.

They add a sparsity constraint (like L1 regularization or KL divergence) so only a few neurons activate, leading to more interpretable feature learning.