A learned value inside a neural network that the model adjusts during training to improve its predictions.