Privacy-preserving machine learning algorithms using lattice-based cryptography

Transforming the machine learning operations in an encrypted domain will gain end-user trust and alleviate their privacy concerns. One of the major challenges in this approach is computational complexity: the existing frameworks do not scale. To overcome the scalability issue, this project proposes a novel approach that exploits the advanced lattice-based Fully Homomorphic Encryption (FHE) such as Torus based FHE to redesign deep learning algorithms to perform data classification in encrypted domain. Use of lattice based FHE enables the algorithm to be secure even after the arrival of quantum computers.

This is an ongoing research project, which started in 2018, and sits at the intersection of the Cyber Security, Trust, Identity and Privacy and Artificial Intelligence, Machine Learning and Data Analytics research themes of the IDT.

One of the most notable outcomes of the project to date is: Y. Rahulamathavan, S. Dogan, X. Shi, R. Lu, M. Rajarajan, and A. Kondoz, “Scalar product lattice computation for efficient privacy-preserving systems”, IEEE Internet of Things Journal, 8(3), pp.1417-1427, 2020.

For further info, please contact Dr Yogachandran Rahulamathavan.