XAI4HiL-ML

The visual inspection and sorting of goods is increasingly being carried out by ML models in industry. Higher accuracy, faster implementation and greater robustness are just some of the advantages of ML models. A core problem of ML models is traceability and reliability.

The XAI4HiL-ML project is dedicated to the question of how Human in the Loop (HiL) approaches can be used to generate better ML models and increase application reliability. At the core of the project are methods for explaining model predictions (eXplainable AI, XAI) and methods for quantifying the uncertainty of the model.

The overarching goal of the project is to define best practices for HiL approaches for the use of ML in visual inspection. The focus is on the user-friendliness of the methods. In particular, the use and optimisation of ML models should be made accessible to people without a background in machine learning with the help of suitable user interfaces and workflows.

The project will thus lay the foundations for the rapid and safe introduction and use of visual inspection systems in commercial operations.