A loss function is a differentiable computation that measures "how incorrect" a prediction is. It is used to calculate the gradient, which in turn guides the direction in which the model updates its parameters at each iteration of the training loop. The output of the loss function is termed "loss," typically computed separately on the training set and validation set (referred to as "training loss" and "validation loss," respectively). Generally, a lower loss value indicates higher accuracy in the model's predictions.
