Privacy-preserving deep learning: algorithms, technologies, and ethics
Date: 2025-03-01
The thesis tackles the algorithmic, technological, and ethical challenges of privacy-preserving ML and DL through Homomorphic Encryption (HE). It proposes methods to adapt models to HE’s computational limits, designs cloud-based platforms for scalability, and examines ethical trade-offs between privacy, transparency, and fairness. Organized in four parts, it spans from foundational concepts to novel HE-compatible architectures, deployment strategies, and ethical analysis.