Gradient Descent

A mathematical technique used in backpropagation to find the direction and amount to change weights to reduce the model's error (loss function) most efficiently. A learning rate hyperparameter affects the size of adjustments.