← 🔁 MLOps & LifecycleNEXT IN MLOPS & LIFECYCLEModel Debugging Methodology→
Core
Drift Detection
Models decay because the world changes. Data drift is a shift in the input distribution (detectable without labels by comparing live features to a training reference with PSI or KS tests); concept drift is a change in the input-to-output relationship (usually needs labels, which often lag). The discipline is monitoring inputs and predictions as leading indicators, alerting on sustained shifts, and triggering retraining. Applied-AI interviews probe it because 'the model was great at launch and quietly got worse' is a top production failure.
a free account unlocks the core curriculum tier · no card
COURSES COVERING THIS TOPIC
No lesson covers this one directly yet. These teach the surrounding topic from the beginning.
RELATED CONCEPTS
PRACTICE THIS IN REAL QUESTIONS
MLOps & ML EngineeringWalk me through concept drift, data drift, and label drift. Which one actually forces a retrain?→MLOps & ML EngineeringHow do you decide when to retrain a production model: on a schedule, or triggered by drift?→MLOps & ML EngineeringWhat should you monitor for an ML model in production (beyond uptime)?→MLOps & ML EngineeringYour ground-truth labels arrive weeks late. How do you monitor the model in the meantime?→MLOps & ML EngineeringHow do you design the triggers and cadence for retraining a fleet of production models?→System Design for AI in ProductionDesign a monitoring system for a fleet of 100+ production ML models.→
COMPANIES THAT ASSUME THIS