Towards a Privacy-by-Design Framework for ML-Enabled Systems

The integration of ML into end-user oriented software systems introduces additional privacy challenges, due to the large amount of data processed and the intransparent nature of ML models, compared to traditional software components. Although techniques for minimizing data usage and increasing privacy exist, they are not yet widely adapted and often create conflicts with system functionality. In this work, we will develop a comprehensive framework that helps developers of ML-enabled systems to specify privacy requirements, monitor their implementation during all phases of software development, and assess tradeoffs between privacy and functionality.