Drift Detection
Domain: MLOps
🎯 Motivation
Detecting when real-world data changes from training data. The model performs worse because the world changed.
📋 Concrete Example
Drift: model trained on 2023 data. In 2025, slang has changed. Model does not understand new slang. Data drift detected. Retrain with new data.
🔗 Analogy
A map becoming outdated - new roads are built, old ones closed. The map (model) no longer matches reality (drift). Retrain with new data.