Software Engineering for AI/ML/DL

Forskningsprojekt, 2019 – 2023

The research is concerned with developing largely automated solutions that support the creation, training and evaluation of ML/DL components in industrial, production-quality systems. Specifically, as, similar to other software components, ML/DL components deployed in industrial practice are continuously retrained, validated using A/B testing and then deployed once the new model proves superior to the previous version, we aim to study the challenges arising from this continuous process of collecting data, improving and retraining the model, validating the newest version of the model and then monitoring and logging to ensure that the model performs according to expectations and integrates well with the other software components as well as the data pipelines established in the software intensive system. As cyber physical systems typically are instantiated many times, one of the challenges that we intend to study is the combination of data and learnings from many instances of systems that operate in, potentially highly diverse, contexts.