Financial institutions can benefit significantly from using machine learning techniques in fraud detection systems. While each institution’s data limitations require collaboration, the data being sensitive-legally and ethically-prevents direct collaboration between them. To resolve this issue, this project will focus on incorporating federated learning and Fully Homomorphic Encryption, which, when achieved with the use of the Concrete library from Zama, allows mathematical operations on encrypted data to some degree. The various ways of combining this notion of encryption with federated learning-while also reviewing various literatures and experimenting with the level of privacy achieved, training speed, accuracy, etc.-will ultimately result in a working privacy-preserving ML framework.

Through this SuperUROP opportunity, I want to learn more about how research is carried out in a longer time frame. I also hope to combine my skills from my previous courses (ML, Software Design, etc.) while learning more about the field of cryptography. I am most excited about being able to build a large system that requires all my skills and being able to improve as a programmer and researcher.