What is a Loss Function

A loss function quantifies how far a model's predictions are from the true target values. It provides the signal that gradient-based optimizers use to update model weights during training.

Without a well-chosen loss function, training has no direction — the optimizer needs a scalar measure of error to know which way to adjust parameters.