Skip to content

Glossary · Model Training

Dropout

A regularization technique that randomly disables a fraction of neurons during each training step, preventing the network from over-relying on any single pathway.

Related pages

More Model Training terms

← Back to the full glossary