Federated Learning (FL) has gained prominence as a privacy-enhancing technology, initially popularized by Google for predicting user keystrokes on smartphones while preserving privacy. FL has expanded to various applications and novel algorithms have been developed to ensure the convergence of FL's global AI model and handle intermittent connectivity in certain scenarios.
However, the strength of privacy protection provided by FL has been called into question by further research. Researchers have identified potential vulnerabilities and threats, such as the curious aggregator threat, susceptibility to man-in-the-middle and insider attacks that disrupt model convergence, and inference attacks that attempt to re-identify data subjects from AI model updates. To address these challenges, the novel Armored Federated Learning (AFL) platform will be developed in this project, aiming to meet the GDPR challenge and enable GDPR compliance while utilizing FL technology.
Another challenge is the accessibility of high-quality datasets for research purposes. Often, datasets of sufficient quality are not available from a single source and must be assembled from subsets owned by different organizations, each with its own access policies. In many cases, these organizations restrict access to entities outside their organization, even more so than required by the GDPR. Furthermore, the mere awareness of the dataset's existence and the ability to contact the data owner pose additional challenges to data accessibility. Overcoming the barrier of accessibility to siloed data in a GDPR-compliant manner would have a significant impact.
Overall, FL and improved accessibility to siloed data in a GDPR-compliant manner hold the potential to revolutionize privacy protection, research capabilities, and the development of critical solutions in various domains.
The goal of TRUMPET project is to research and develop novel privacy enhancement methods for Federated Learning, and to deliver a highly scalable Federated AI service platform for researchers, that will enable AI-powered studies of siloed, multi-site, cross-domain, cross-border European datasets with privacy guarantees that exceed the requirements of GDPR. The generic TRUMPET platform will be piloted, demonstrated and validated in the specific use case of European cancer hospitals, allowing researchers and policymakers to extract AI-driven insights from previously inaccessible cross-border, cross-organization cancer data, while ensuring the patients’ privacy. The strong privacy protection accorded by the platform will be verified through the engagement of external experts for independent privacy leakage and re-identification testing.
A secondary goal of the project is to research, develop and promote with EU data protection authorities a novel metric and tool for the certification of GDPR compliance of Federated Learning implementations.
