Glossary › weights
GLOSSARY
weights
appears in 1 paper titles
Definition
The learned parameters of a model — in a neural network, the numbers scaling each connection between units. They are the model in any practical sense: releasing weights is what open-weight publication means, and loading them is what running a trained model consists of. Distinguish them from hyperparameters, which are set by the practitioner rather than learned, and from sample weights in statistics.