Former projects

Artificial intelligence needs data - ideally a lot of it - in order to train for pattern recognition. Using raw data sometimes raises privacy issues in this regard, as it may contain sensitive or competition-sensitive information. For example, sensitive patient information can be found in medical data or valuable production information in industrial data. However, this sensitive data is often not needed by the AI at all, as the patterns in the data become visible anyway. Homomorphic Machine Learning addresses precisely this challenge of how to train AI algorithms on anonymized data - with results similar to those achieved when training with raw data.

The SMILe ("Secure Machine Learning Applications with Homomorphically Encrypted Data") project explores new ways and solutions to harness the potential of sensitive data for machine learning using homomorphic encryption. To this end, existing platforms are extended to create collaborative-creative ecosystems in which data privacy and security are ensured as well as the previously untapped potential of data can be exploited.

VRVis contributes its visualization and visual analytics expertise to the SMILe project. VRVis visualizes the differences in the AI results with raw data as well as anonymized data and contributes through visual exploration to make the black box character of the machine learning models "observable" as well as the effect of encryption on the precision of pattern recognition comprehensible, thus establishing trust in the AI trained on homomorphically encrypted data.