Report: State of the Art Solutions for Privacy Preserving Machine Learning in the Medical Context
This addresses privacy challenges for medical data holders and researchers, but it is incremental as it reviews existing solutions rather than proposing new ones.
The paper evaluates cryptographic mechanisms for privacy-preserving machine learning in medical contexts, focusing on homomorphic encryption, multiparty computation, and trusted execution environments to enable data processing without violating privacy under regulations like GDPR.
Machine Learning on Big Data gets more and more attention in various fields. Even so privacy-preserving techniques become more important, even necessary due to legal regulations such as the General Data Protection Regulation (GDPR). On the other hand data is often distributed among various parties. Especially in the medical context there are several data holders, e.g. hospitals and we need to deal with highly sensitive values. A real world scenario would be data that is held in an electronic patient record that is available in many countries by now. The medical data is encrypted. Users (e.g. physicians, hospitals) can only decrypt the data after patient authorization. One of the main questions concerning this scenario is whether it is possible to process the data for research purposes without violating the privacy of the data owner. We want to evaluate which cryptographic mechanism - homomorphic encryption, multiparty computation or trusted execution environements - can be used for this task.