Review Weights

Weights are learned parameters that determine how strongly signals influence one another inside a model.

During training, the learning process repeatedly adjusts weights. Together, those values encode patterns the model uses when it processes a new input.

Weights are central to how trained models behave and to questions about distribution, modification, reproducibility, and whether a model can run outside a vendor's service.

Training a neural network changes its weights so features associated with a target category contribute more appropriately to the final prediction.

A weight is not a readable fact stored in a single location. Useful behavior usually emerges from many values operating together.

Weights determine how much each learned signal counts.

* [[parameters|Parameters]] * [[neural-network|Neural Network]] * [[open-weights-vs-closed-weights|Open Weights vs. Closed Weights]] * [[training-data|Training Data]]

* [Google for Developers, *Machine Learning Glossary: Weight*](https://developers.google.com/machine-learning/glossary#weight) — technical definition of weights as learned coefficients in a model. * [Stanford CRFM, *On the Societal Impact of Open Foundation Models*](https://crfm.stanford.edu/open-fms/) — explains the practical significance of widely available model weights for use, inspection, customization, and governance.

* Connect internal values to distribution choices in [[open-weights-vs-closed-weights|Open Weights vs. Closed Weights]].