For Heads-of · Practitioner
Performance Regression
A decline in model performance over time, typically caused by shifts in the data, concept drift or changes in the environment the model operates in.
- performance regression
- degradation
- monitoring
- maintenance
Causes
Real-world data drifts away from the training data, user behaviour changes, new kinds of input appear, a provider updates a hosted model, or feedback loops change the environment the model operates in.
Detection
Continuous monitoring of performance metrics shows a regression, and breaking the metrics down by group, time period or input type points to the cause. AI system monitoring and logging is the control.
Response
Options include retraining on current data, updating features or adjusting decision thresholds. A serious regression may need immediate replacement or a fall-back to human decision-making through AI incident response and rollback. Regressions nobody notices are the register's AI model drift and degradation.
Related Terms
Model Drift
A change in the relationship between a model's inputs and the environment it was fitted to, such that the model's assumptions no longer hold. The model itself is unchanged; the world it was trained on has moved.
Model Monitoring
Continuous observation of a deployed model's inputs, outputs, performance and behaviour, to detect degradation, drift, misuse or security anomalies.
Model Versioning
The practice of recording which model, trained how and on what, was deployed when, enabling comparison, rollback and reproducibility. For systems built on models the organisation did not train, it extends to versioning the surrounding assembly, since that is the only part it controls.
Performance Metric
A quantitative measure of how well an AI model or system performs a task, such as accuracy, precision, recall, F1 score or a domain-specific measure.
Related risks and controls
AI model drift and degradation
The register entry for the degradation this measures.
Untested AI failure scenarios
A regression in a category nobody tests is one nobody detects.
Disparate impact monitoring gap
Fairness regresses independently of accuracy, and separately from it.
AI system monitoring and logging
Supplies the measurement without which a regression is an anecdote.
AI incident response and rollback
What a critical regression triggers.