SHIELD: Secure HomomorphIc Encryption for privacy-preserving machine Learning in the presence of siDe-channel attacks

Fully Homomorphic Encryption (FHE) has emerged as a vital privacy-enhancing technique for privacy-preserving machine learning (PPML) applications.

As the demand for PPML solutions rises in various sectors, such as healthcare, finance, and social media, ensuring the security of FHE schemes becomes crucial.

This four-year project aims to address the challenges in FHE security by investigating the mathematical foundations and physical implementations of FHE, focusing on mitigating side-channel vulnerabilities across multiple platforms.The research encompasses decreasing the concrete complexity for solving the Learning with Errors (LWE) instances utilized in FHE schemes, assessing and enhancing the security of FHE schemes like TFHE and CKKS, and securely implementing open-source FHE libraries.

By identifying and addressing potential security breaches proactively, the project seeks to deepen our understanding of FHE security, substantially impacting the security of PPML applications.Significant to the Swedish industry, this project highlights the need for rigorous testing of FHE implementations before deploying them in real-world applications, ultimately promoting FHE adoption throughout Sweden.

As FHE transitions from a research frontier to practical applications, the project´s timely initiation ensures a solid foundation for future FHE security research and guarantees the security of PPML applications in the long run.