Regularization
Domain: Evaluation
🎯 Motivation
Techniques to prevent overfitting by penalizing complexity. The model learns simpler, more generalizable patterns.
📋 Concrete Example
L2 regularization: add penalty for large weights. Model learns simpler patterns. Training accuracy: 95% -> 88% (slightly worse). Test accuracy: 60% -> 82% (much better). Less overfitting.
🔗 Analogy
A coach telling an athlete ‘do not overtrain’ - too much specific training causes injury (overfitting). Regularized training keeps the athlete healthy and adaptable.