Cryptographic building blocks for privacy preserving AI

Purpose and goal:

The project successfully enhanced AI competence in privacy protection and secure data sharing at the Swedish target organization. It developed technical skills in post-quantum cryptology and cryptographic primitives, deepened understanding of encrypted neural networks’ scaling properties, which can promote knowledge dissemination for the benefit of Swedish industry and the public sector. Findings from the collaboration were disseminated at two international conferences.

Expected results and effects:

The project increased competence in post-quantum cryptology and cryptographic primitives, and deepened understanding of encrypted neural networks. It facilitated knowledge dissemination through conferences and workshops and led to continued collaborative efforts. This can contribute to enhancing the ability of both the Swedish industry and public sector to develop secure, resource-efficient AI systems and implement trustworthy AI applications, fostering innovation and trust in AI across society.

Approach and implementation:

The project began with training on cryptographic building blocks, followed by collaborative development of experiment source code. On-site visits and workshop participation facilitated hands-on learning and knowledge exchange. The final phase involved compiling results, source code, and documentation.

This structured approach enabled skill development and laid the groundwork for ongoing collaboration, including a Master’s project to further develop the work and promote knowledge transfer to junior engineers.