Prof. Dr. Zekeriya Erkin

(Delft University of Technology)
hosted by Secure Decentralized Systems group of Stefanie Roos

"ML meets MPC: Financial Crime Detection with Privacy"

Privacy-enhancing technologies, particularly MPC, have been questioned about their practicality. Last summer, a competition was organized by the USA and UK governments (U.S.-U.K. PETs Prize Challenges: link: https://www.drivendata.org/competitions/group/nist-federated-learning/ ), where privacy researchers were tasked to solve the fraud detection problem using machine learning algorithms while providing privacy guarantees to banks and their users. For this purpose, we developed the cryptographic part of the solution, where the result is a custom protocol that allows payment providers to check whether a user’s credentials match the credentials in a bank’s database without the bank learning any information about the user. At the same time, the payment provider never has access to the bank’s database. These seemingly-impossible constructions are made possible using a particular type of encryption called homomorphic encryption. The competition results show that privacy-enhancing technologies are fast enough to be practically deployable. They provide solutions for problems that are otherwise hard to address without exposing private data.

Time:	Thursday, 13.06.2024, 13:30
Place:	room 36/432

Termin als Datei downloaden und in den Kalender importieren.