A mathematical function that measures the difference between predicted and actual values, guiding the model learning process. Different tasks use different loss functions like mean squared error or cross-entropy.