Mitigation¶
Machine learning models in production environments have a complex lifecycle; maintaining the predictive value of these models requires a robust and repeatable process to manage that lifecycle. Without proper management, models that reach production may deliver inaccurate data, poor performance, or unexpected results that can damage your business’s reputation for AI trustworthiness. Lifecycle management is essential for creating a machine learning operations system that allows you to scale many models in production.
| Topic | Description |
|---|---|
| Retraining | Schedule and manage model retraining jobs for deployed models. |
| Humility | Configure humility rules so models flag uncertain predictions in production. |
| Challenger models | Register and compare challenger models against champion deployments. |
| Model replacement | Replace the model behind an existing deployment without creating a new one. |