Loss

Domain: Evaluation

🎯 Motivation

The model’s error during training. Lower loss = better fit to training data. The training process minimizes loss.

📋 Concrete Example

Training loss: starts at 4.5, drops to 0.3 after 10 epochs. The model is learning. If loss stops dropping, the model has converged. If loss goes up, overfitting.

🔗 Analogy

A golfer’s score - lower is better. They practice to reduce their score (loss). Training minimizes loss like practice improves a golf score.

Related Concepts

• Metric
• Training Loss
• Cross-Entropy Loss