Model drift happens when real-world data changes and an AI becomes less accurate.
It is like using last summer's weather app in a snowstorm. The forecast is still trying its best. Bless its little pixels.
People watch for drift in fraud checks and recommendations. They update or retrain the model with new data.
Model regression
Model drift comes from real-world change. Model regression often follows an update.
Continual Learning
Continual Learning uses new data to help a model adjust over time.
Model Calibration
Drift can make a model's confidence score less trustworthy.