Privacy-preserving Machine Learning

Summary:

Machine learning (ML) algorithms are often trained on sensitive data that may not be shared without certain privacy guarantees in place. This is particularly relevant for situations where multiple organisations (i.e., data owners) would like to collaborate with each other by sharing their data but are concerned about privacy. One of the strongest guarantees can be provided through the encrypted processing of the data without ever decrypting it. The common techniques in this context are multi-party computing, masking, homomorphic encryption, garbled circuits and differential privacy. In this line of research we are interested in the development of variants of ML algorithms that can work on encrypted data by using cryptographic constructs, using privacy-enhancing technologies as an enabler.

Participants

• Fatih Turkmen
• Ali Reza Ghavamipour

Students:

• Sina Rouzbahani: MSc Intern rivacy-preserving Logistic Regression training framework using Multi-Party Computation, 2021.

Publications

• Ali Reza Ghavamipour, Fatih Turkmen and Xiaoqian Jiang, Privacy-preserving Logistic Regression with Secret Sharing, 2021, Preprint, (Accepted).