Secure Federations: Addressing the Security Challenges in Federated Learning and Privacy-Preserving AI
Elizabeth Ango Fomuso Ekellem · 2023
Federated Learning (FL) and Privacy-Preserving AI present a promising frontier for fostering collaborative machine learning while upholding data privacy at individual levels. Yet, these advancements are not without challenges, notably opening new corridors for adversarial incursions. In this paper, we delve into the intricate security impediments inherent within FL and privacy-centric AI. We cast a spotlight on dominant attack vectors, notably model poisoning, and data injection, with an emphasis on healthcare applications [1][4]. Through a meticulous exploration of their operational modalities, repercussions, and viable defenses, we craft security protocols tailored for FL infrastructures [2]. Venturing further, we evaluate the delicate balance between ensuring model robustness and sustaining optimal performance, underpinned by rigorous empirical studies. Our discourse accentuates not just the vulnerabilities lurking within FL, but also charts a structured path towards fortifying distributed machine learning ecosystems [6].