Concept
Dropout
Dropout is a regularization technique that randomly zeroes a fraction of a layer's activations during each training step. Introduced by Srivastava et al.
The rest of “Dropout” is a premium feature: every concept in the library gets a precise, practitioner-focused write-up like this one, cross-linked straight from the paper summaries that use it.
Log in to unlock→