Glossary · IIoT, data and AI
Model deployment
Also known as: Model serving, Model rollout
German: Modellbereitstellung
In machine learning operations, model deployment is the process of making a trained and validated model available for use in its target environment, such as a cloud service, an edge device or a controller, including packaging, installation, configuration, integration and release.
- IIoT
- AI
In one sentence
Model deployment makes a trained, validated model available in its target environment, such as a cloud service, edge device or controller.
Example
A validated defect detection model is packaged as a container, deployed to the line's edge device in shadow mode for a week and then activated.
How it applies
- Engineering: Common strategies include shadow mode (the model runs but does not act), canary rollout to one line first, and blue-green switching with a fast rollback path (Rollback capability).
- Commissioning: Verify that inputs in the target environment match those used during training and validation; otherwise Training-serving skew can degrade results immediately.
- Documentation: Record which model version is deployed where, with which configuration, who approved it and when. Update operator instructions if the model changes what operators see or must do.
Model deployment vs. model release
Deployment is the technical installation. The release is the decision that the model may be used for its intended purpose (Release decision). A model can be deployed for testing without being released for production use.