Dr. Yogachandran Rahulamathavan
Loughborough University London, United Kingdom
Yogachandran Rahulamathavan is a Senior Lecturer and the Programme Director for MSc Cyber Security and Data Analytics at Loughborough University’s London Campus in the UK. Yoga obtained his PhD degree from Loughborough University in mathematical optimisation techniques for information processing in 2012. His research interest is on developing novel security protocols to advance machine learning techniques to solve complex privacy issues. Currently focussing on post-quantum encryption techniques to develop privacy-preserving machine learning algorithms. Currently, Dr Rahul coordinates a UK-India project between Loughborough University London, IIT Kharagpur, India and City, University of London. He is a Senior Member of IEEE and an Associate Editor for the IEEE Access journal.
Title: "Hide-and-Seek: Machine Learning in Encrypted Domain"
Abstract: Machine Learning models were built using a huge amount of high-quality and application-specific data. Even though the machine learning models can only be trained at places where the data is available, anyone can use the trained model for classification tasks via the Internet. While it sounds revolutionary, the trained ML models are not readily available to users in healthcare, finance, or marketing due to privacy issues. Users do not want to share their sensitive data with service providers due to a lack of trust. Simply encrypting the data only protects them during storage and transmission. Researchers and industries are developing novel techniques known as privacy-preserving techniques to process the data in an encrypted domain to tackle the privacy issue. Homomorphic encryption plays a key role in developing privacy-preserving machine learning algorithms. While homomorphic properties exist in traditional cryptographic schemes such as RSA, this talk will focus on fully homomorphic encryption from lattice-based cryptography. We will also go through the state-of-the-art works, challenges and future trend in this domain.