MLOps Architecture for Training Data Lineage

Avatar von manuel.pasieka


Implementing continuous training of Machine Learning Models is hard. Keeping track how and on what data the latest model was trained is even harder.

As AI Consultant I designed a MLOps Architecture to track the training data sets used for image analysis pipelines. Enables training data lineage, which describes what training data was used to train which model and is an integral part in automatic re-training Machine Learning Pipelines to ensure quality and stability of the system.

This provides control and ensures robustness. Making it possible to understand where changes in the behavior or performance of models come from and where to start investigating if problems emerge.