5. května 2023 In April 2023, we have finished a European research project called PRIVILEGE (PRIVacy and homomorphIc encryption for artificiaL intElliGencE) that started in 2020 and was funded by Preparatory Action on Defence Research (PADR) a precursor programme of the European Defence Fund (EDF). CESNET collaborated in this project with three other organizations from European countries: Thales (as a project coordinator) from France, CEA from France, and Intracom Defense from Greece. The project was focused on emerging topics related to data privacy in the context of collaborative machine learning.
Building a common machine learning model by sharing data among multiple parties can be beneficial in many applications. However, in some use cases, the data may be too sensitive, or even classified, and can not be shared, making traditional ways of collaborative learning impossible. Therefore, the project focused on securing the process of collaborative learning such that it is possible to create a common model combining knowledge of multiple parties without the risk of disclosing private data to an untrusted party.
The main outcome of the project is the PRIVILEGE framework that helps to utilize secure collaborative learning in particular applications using a combination of Federated Learning (FL) or Private Aggregator of Teachers Ensemble (PATE) methods with Fully Homomorphic Encryption (FHE), Multi-Party Computation (MPC), or Verifiable Computation (VC). CESNET contributed to the international consortium as a use case provider with our network security expertise, in particular a machine learning-based method for prediction of cybersecurity events.