Research Projects

The Internet has revolutionized our lives, allowing for collaboration among people worldwide who have never met or interacted with each other. This has resulted in many benefits, e.g., recommendation systems used in e-commerce and collaborative editing tools like Google Docs and Github. These benefits come at the cost of the privacy of the individuals using these services. There is tension between the Internet's benefits and individual privacy. Secure Multiparty Computation (MPC), Distributed Oblivious Random Access Memory (DORAM), Private Information Retrieval (PIR), and Zero-Knowledge Proofs (ZKPs) are some cryptographic techniques that help loosen this tension, i.e., keep the benefits of the Internet while not compromising on privacy. This project proposes to leverage these cryptographic techniques to allow for privacy-preserving collaborations in three key areas: recommendation systems, collaborative editing, and financial fraud detection The state-of-the-art privacy-preserving recommendation systems use PIR to download the data privately, and MPC atop those PIR queries to provide recommendations. Current privacy-preserving recommendation systems use a PIR scheme that relies on strong non-collusion assumptions, often requiring no two parties to collude. We propose to relax this assumption by using alternative PIR protocols like Goldberg’s IT-PIR, which uses Shamir’s secret sharing. However, implementing MPC atop Goldberg’s PIR is less efficient. This project will develop novel MPC protocols atop Goldberg's IT-PIR to provide recommendations based on users' consumption history. We will also integrate differential privacy (DP) with MPC to hide the inherent leakage caused by running the recommendation algorithm. Integrating MPC with DP is a very challenging task because sampling from a random distribution in MPC is known to be a difficult problem. This project proposes to find novel MPC protocols to sample from a random distribution and thus facilitate the integration of DP with MPC. Next, the project will develop privacy-preserving collaborative editing tools. A collaborative editing tool with privacy requires two main properties: i) allowing users to write data anonymously, which can be achieved using DORAMs, and ii) only allowing authorized users to write or edit data. To achieve authorization atop anonymous writing, we will use Zero Knowledge Proofs. Our system will also support the transfer of editing rights privately, thus allowing multiple people to write the same document without conflicts. Finally, the project will tackle financial fraud by allowing different institutions to run ML algorithms on their private data using MPC to detect fraudulent activities. A challenge in privacy-preserving systems is their limitation to proof-of-concept implementations. The project will bridge this gap by translating theoretical constructs into practical, deployable software solutions, ensuring real-world applicability and impact.