Dr. Sadegh Riazi receives Jacobs Graduate Council Award (which is given to one graduate student from all six engineering schools within Jacobs school) and William S.C. Chang Best Ph.D. Dissertation Award (a Electrical and Computer Engineering Department award) for his dissertation, titled "Towards a Private New World: Algorithm, Protocol, and Hardware Co-Design for Large-Scale Secure Computation".
Dr. Riazi has been working with Prof. Koushanfar in the Electrical and Computer Engineering Department. His research focus is a new set of technologies that enable computation on encrypted data.
Advancing science in several research domains, including Artificial Intelligence (AI), requires access to a large volume of data. However, the data often contains sensitive information about individuals and cannot be used freely. Moreover, in many scenarios, the data is not owned by a single central entity and is held by different parties who do not wish to share their data.
Can we continue to advance science and technology using the data that we can't have direct access to?
Now, more than ever in the past, there is a need to consolidate data privacy requirements and the ability to utilize data. Traditional encryption mechanisms such as Advanced Encryption Standard (AES) allow data to be securely communicated from point A to point B. However, the encrypted (unintelligible) data cannot be used for any computation and has to be decrypted once it's needed.
Computer scientists, for many decades, have studied various cryptographic ways to enable computation on encrypted data without access to the decryption key. The proposed solutions come with enormous computational overhead, prohibiting them from being used in practice.
Dr. Riazi's Ph.D. dissertation proposes novel methodologies that enable computation on encrypted data at an unprecedented scale and performance. His thesis introduces new algorithms, protocols, and hardware architectures to increase the efficiency and scalability of computing on encrypted data by one to two orders of magnitude in each of these categories. Such advances open new doors for privacy-preserving machine learning, secure cloud computing, and personal genomics while protecting individuals' privacy.
A Real-World Example: Privacy-Preserving AI-based Medical Diagnosis
There has been significant progress in AI-based medical diagnosis. In this scenario, hospitals can send the medical record of patients to third-party companies, where AI models are used to process the data and make an automated medical diagnosis. One of the major obstacles in realizing this scenario is data privacy. Several privacy laws, such as HIPAA, require the protection of health records and do not allow third-party companies to have access to the raw patients' data.
Dr. Riazi proposes a new framework in which hospitals can encrypt medical records and allow AI companies to process the data without learning and having access to the raw data. The proof-of-concept implementation of the system for four diagnosis tasks, i.e., breast cancer, diabetes, liver disease, and Malaria, shows less than one second processing time while guaranteeing the privacy of the patients.
