Abstracts

More and more of today's applications use AI systems developed through machine learning. The decisive factor here is not only that there is enough or even quite a lot of data available for learning, but also that this data is of sufficient quality. But what exactly is meant by quality? Many of the aspects of data quality that are known in the context of machine learning are rather qualitative in nature. How do such aspects become high quality? Can they also be made measurable? The talk will shed light on data quality issues and describe the challenges involved: How does one assess qualitative features as objectively as possible, what influence do these have on the quality of the model, and how can the explainability of AI models help to improve the data quality? For some of these questions, the talk will show approaches to solutions that were developed in the context of a scientific study. The presentation is aimed at software developers and testers, but also test managers who deal with artificial intelligent systems and for those who value data quality highly.