The project developed a system for deploying machine learning models in production using the Fraunhofer Edge Cloud. It features automated deployment, training, and tracking of models using MLflow for lifecycle management and GitLab ci/cd for workflow automation. The system also supports automatic retraining based on performance degradation. Deployment strategies vary by use case, with direct model delivery through SSH for real-time systems, and encapsulation in a WebAPI for others.
| Topic Fields | |
| Published | 2021 |
| Involved Institutes | |
| Project Type | ICNAP Research/Transfer Project |
| Responsibles |
This project focused on the development of a concept for the deployment of machine learning models on production systems from the Fraunhofer Edge Cloud. For this purpose, in addition to automated deployment, the possibility of automated training and tracking of models and their artifacts was achieved. The MLflow platform is used to manage the ML lifecycle, including experiments, reproducibility, deployment, and a central model registry. Automation of the entire workflow is performed using the ci/cd infrastructure of the GitLab instance running at Fraunhofer. Within this framework, the concept was extended to include automatic retraining, which is based on the degradation of the model in production. This involves analyzing the input and output data of the running model to monitor its drift and retrain when the model is no longer appropriate. Each use case depends on a different strategy for deployment, for those that require a model delivery directly, as in the case of a realtime system a ssh connection is made to the pipelines for model delivery, in other cases the model is encapsulated in a WebAPI to provide information in an isolated way from the system.
Contact us to get in touch! With a membership, you’ll gain full access to all project information and updates.
© Fraunhofer 2025
