Associated Projects

The “KI-AIM2” project builds on successful prior work and a developed platform (Cinnamon) for the depersonalization of medical data and extends it in key areas. The platform will be expanded with a component for free-text anonymization and synthesis, which account for 80% of stored medical information. Interoperability will be investigated so that the platform can be integrated into existing health information systems and (partially) automated anonymization pipelines can be implemented in practice. In addition, innovative models for the integrated quantification of privacy risks and the realism of output data for anonymized and synthesized datasets will be developed.

Through innovative and practical anonymization methods, the project aims to simplify access to medical data for research purposes. The developed platform will be released as open-source software, and a large user community will be established. The increased availability of demonstrably anonymous medical data is intended to help enable the development of AI models for a wide range of medical applications. On this basis, entirely new applications in the field of connected health data analysis as well as personalized medicine are conceivable.

Ensuring anonymity guarantees is of crucial importance for the implementation of new data-driven applications. Classical anonymization and pseudonymization methods such as so-called k-anonymity or differential privacy are already used in practice, but they prove unsuitable for real-world application scenarios. One reason for this is that, in real applications, more complex data types are increasingly replacing simple numerical data types, such as sequences of location data or hierarchical data structures that represent physical or logical conditions. In addition, existing methods are unable to support communication and architecture models relevant to enterprise environments, meaning that they cannot be integrated into established software stacks. This is particularly true for so-called stream processing models, in which large amounts of data continuously pass through a processing chain without being permanently stored. Established metrics and methods for estimating and balancing processing effort and information loss associated with different anonymization methods are also lacking.

The aim of the project “Ensuring Anonymity Guarantees in Enterprise Streaming Applications (GANGES)” is to overcome the aforementioned obstacles by developing practice-oriented anonymization methods that are aligned with real-world requirements for anonymity guarantees. To this end, new methods for stream anonymization will be developed based on concrete use cases from the energy and building management sectors as well as existing baseline methods. Particular consideration will be given to complex data types commonly used in practice, which pose specific anonymization challenges. The developed methods will be integrated with minimal effort into established open-source streaming frameworks. Furthermore, new approaches for measuring and experimentally determining processing effort and the remaining utility of data after anonymization will be developed. Based on this, the methods and integration technologies will be continuously evaluated and refined throughout the development process.