Concept
Co-adaptation
Co-adaptation is when hidden units in a neural network learn to work only in the presence of specific other units, forming brittle coalitions that fix each other's mistakes rather than each computing an individually useful feature. Think of it like overfitting, except the…
The rest of “Co-adaptation” 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→