Validation Loss
Domain: Evaluation
🎯 Motivation
Loss measured on validation data (not used for training). If training loss drops but validation loss rises, the model is overfitting.
📋 Concrete Example
Training loss: 0.3 (low). Validation loss: 2.1 (high and rising). The model memorized training data but cannot generalize. Overfitting detected.
🔗 Analogy
Practice exam (training) score: 95%. Real exam (validation) score: 60%. You memorized practice answers but cannot solve new problems. Validation loss detects overfitting.