College of Engineering Unit:
A machine learning operationalization project focused on Intel’s image services. Provides infrastructure for deploying machine learning models to detect defects in welds.
Intel’s machine learning models are distributed to customers via an open-source framework called MLflow. This project expands the existing framework to implement automated model drift detection and correction.
Model accuracy is tracked and updated on a periodic schedule. Models with decreased accuracy over time are labeled as “drifting” and discarded from the model registry. Corrections are also made to the data used to train the “drifted” models.