Weights are the parameters of a network that are adjusted during training. The weights are updated using gradients which are derivatives of the Loss Function with respect to the weights.