Interests and expertise

Machine learning (ML) is used in various domains including healthcare, finance, and retail to train predictive models for use and in recent years ML-as-a-service platforms have gained a lot of popularity. Using large datasets from various sources helps improve the training accuracy of these models, but such massive data collection poses serious privacy concerns.

My work focuses on privacy-preserving training of machine learning models in two settings i.e. multi-party computation and federated learning settings. For the multi-party computation (MPC) setting, current works use expensive protocols e.g. garbled circuits for computing non-linear activation functions and their derivatives, as it is technically challenging to compute these functions in an MPC setting. Similarly, for the federated learning setting, data privacy comes at a cost of training inefficiency. My work focuses on designing and implementing new protocols for privacy-preserving machine learning training that are also practical and efficient to be used in the real-world.

Commentary

At Supra, I have been involved in various research problems including Distributed Key Generation (DKG) in particular I have worked on Class group DKG. I have also been involved in the implementation of various cryptographic protocols and libraries.

Selected research publications