TRUstworthy Multi-site Privacy Enhancing Technologies

近年来，联合学习(FL)已经成为一种革命性的隐私增强技术，并因此迅速扩展到其他应用。然而，进一步的研究给FL提供的隐私保护力度蒙上了一层阴影。研究人员指出的潜在漏洞和威胁包括奇怪的聚合器威胁；对中间人和内部攻击的敏感性，这些攻击扰乱了全局和局部模型的收敛或导致收敛到虚假的最小值；以及最重要的是，旨在从FL的AI模型参数更新中重新识别数据主体的推理攻击。Trumet的目标是为联合学习研究和开发新的隐私增强方法，并为研究人员提供一个高度可扩展的联合AI服务平台，使基于AI的研究能够对孤立的、多站点的、跨域的、跨境的欧洲数据集进行研究，并提供超出GDPR要求的隐私保障。通用小号平台将在欧洲癌症医院的特定用例中进行试点、演示和验证，使研究人员和政策制定者能够从以前无法访问的跨境、跨组织的癌症数据中提取人工智能驱动的见解，同时确保患者的隐私。该平台给予的强大隐私保护将通过聘请外部专家进行独立的隐私泄露和重新识别测试来验证。第二个目标是与欧盟数据保护当局一起研究、开发和推广一种新的指标和工具，用于认证FL实施的GDPR合规性。该联盟由9个跨学科合作伙伴组成：3个研究机构、1个大学、3个中小企业和2个临床合作伙伴，他们拥有丰富的经验和专业知识，以保证活动的正确开展和成果的实现。

In recent years, Federated Learning (FL) has emerged as a revolutionary privacy-enhancing technology and, consequently, has quickly expanded to other applications. However, further research has cast a shadow of doubt on the strength of privacy protection provided by FL. Potential vulnerabilities and threats pointed out by researchers included a curious aggregator threat; susceptibility to man-in-the-middle and insider attacks that disrupt the convergence of global and local models or cause convergence to fake minima; and, most importantly, inference attacks that aim to re-identify data subjects from FL’s AI model parameter updates. The goal of TRUMPET 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 is to research, develop and promote with EU data protection authorities a novel metric and tool for the certification of GDPR compliance of FL implementations. The consortium is composed of 9 interdisciplinary partners: 3 Research Organizations, 1 University, 3 SMEs and 2 Clinical partners with extensive experience and expertise to guarantee the correct performance of the activities and the achievement of the results.