Towards Privacy-Preserving Machine Learning in Sovereign Data Spaces: Opportunities and Challenges

The world of big data has unlocked novel avenues for organizations to generate value via sharing data. Current data ecosystem initiatives such as Gaia-X and IDS are introducing data-driven business models that facilitate access to diverse data sources and automate data exchange processes among organizations. However, this also poses challenges for organizations and their customers in preserving control over their own data. This paper provides an overview of the extension requirements on current usage control concepts in data spaces through technical means to augment data privacy guarantees. Our analysis clarifies the deficiencies regarding privacy within the realms of data sovereignty and sovereign data spaces, as well as the risks and opportunities associated with the application of machine learning on sensitive data. This work identifies promising foundational elements and presents areas of research for the integration of privacy-enhancing technologies into usage control for remote data science.