Privacy-first Information Algorithm Designs

Speaker: Mingxun Zhou
Carnegie Mellon University

Title: Privacy-first Information Algorithm Designs

Date: Thursday; 6 February 2025

Time: 4:00pm - 5:00pm

Venue: Rm 2306 (via lift 17/18), HKUST

Abstract:

Information infrastructures — such as search engines, data collection systems, and analytics pipelines — form the backbone of critical applications like media platforms, machine learning services, and financial systems. Recent advancements in AI have propelled these systems to unprecedented levels of utility, scale, and ubiquity. However, users are often forced to sacrifice their privacy rights when interacting with these applications, leading to potential information leakage and even threats to personal safety. Privacy-first algorithm design addresses these challenges by prioritizing data privacy and security as foundational principles, creating a win-win scenario in these information infrastructures: users retain essential privacy protections, while providers can deliver high-quality services and responsibly leverage data to enhance their offerings.

In this talk, I will present our research in fundamental areas of privacy-first algorithm design, with a focus on Private Information Retrieval (PIR) and Private Information Collection (PIC). In the PIR direction, I will discuss our work on building privacy-preserving retrieval and search engines, including a series of research contributions (Eurocrypt 2023, S&P 2024, Eurocrypt 2024). In the PIC direction, I will introduce Conan (CCS 2024), a novel privacy-preserving information collection protocol that guarantees full anonymity for honest participants while providing robust compliance checking to detect and defend against malicious actors.

Finally, I will highlight exciting open questions and future directions, such as Privacy-Preserving Machine Learning (PPML) and regulatory challenges in Decentralized Finance (DeFi), making a call for future collaboration across academia, industry, and policymakers to address the critical data security and privacy challenges.

Biography:

Mingxun Zhou is now a PhD candidate in the Computer Science Department at Carnegie Mellon University, advised by Prof. Elaine Shi and Prof. Giulia Fanti. He is interested in privacy-preserving and compliant algorithm design for real-world applications including information retrieval and searching, data aggregation, machine learning and AI, etc. He has 10 research papers (9 as the primary author) published at top-tier academic conferences (e.g., IEEE S&P, CCS, Eurocrypt, NDSS), and some of his past works have been deployed in industrial production systems. He also received the CyLab Presidential Fellowship (2023) at CMU. See his homepage for more information: https://cs.cmu.edu/~mingxunz.