Hardware-Assisted Machine Learning on Encrypted Data in the Real World
Dr. Evgenios Kornaropoulos
KEY INTERESTS
Computer security; Applied cryptography; Analysis and design of efficient encrypted systems and algorithms; Artificial intelligence and machine learning
APPOINTMENTS/AFFILIATIONS
Assistant Professor, Department of Computer Science, George Mason University
ACADEMIC DEGREES
BS, Computer Science, University of Crete
MS, Computer Science, University of Crete
MS, Computer Science, Brown University
PhD, Computer Science, Brown University
HARDWARE-ASSISTED MACHINE LEARNING ON ENCRYPTED DATA IN THE REAL WORLD
This research focuses on the development of techniques for applying machine learning (ML) models on encrypted data using a Trusted Execution Environment (TEE). The advancement of, and increased accessibility to, prototype scalable ML models has led to widespread concern about the lack of transparency in leveraging users' sensitive data, undermining public trust in these systems. TEE technology is a hardware-based answer to the challenge of executing software on a remote computer owned and maintained by an untrusted party, protecting the confidentiality and integrity of data while a computation is being performed upon it. This project identifies ML models, and accompanying benchmark datasets, to develop doubly-oblivious algorithms, deployed using a TEE, to address concerns in the given machine learning models.