Hao Han

Professor

College of Computer Science and Technology
Nanjing University of Aeronautics and Astronautics
29 Jiangjun Boulevard
Nanjing 211106, China

Office: College Building #236
Email: hhan@nuaa.edu.cn

Short Biography

Dr. Hao Han is a Professor in the College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics (NUAA). He received his Ph.D. degree from the College of William and Mary in 2014, advised by Prof. Qun Li. Prior to joining NUAA, Dr. Han worked at IBM China Research Laboratory, Microsoft Research Asia, and Intelligent Automation Inc., US. He currently leads the Security of Smart Systems Lab (S3Lab) at NUAA. He received the Jiangsu Distinguished Professor（江苏省特聘教授）, the Six Talent Peaks Project of Jiangsu Province （江苏省六大人才高峰）, and the ACM Nanjing Rising Star Award. Dr. Han's research has been mainly focused on System Security in the the practice of protecting intelligent systems from unauthorized access, tampering, or data leakage. He is a member of CCF, IEEE and ACM.

Publication

*Tips: use doi2bib to convert DOI to BibTeX entry
Journal Articles
Conference Articles

Teaching

Undergradute-Level Courses

- Operating Systems（中文）, Spring 2019 - Present
- Principle of Computer’s Composition (English), Fall, 2019 - 2020

Graduate-Level Courses

- Security in Computer System (English), Fall 2020 - 2023

Awards of Mentoring Students

- 2025，“中国网谷-华为杯”中国研究生网络安全创新大赛，全国二等奖
- 2024，“中国网谷-华为杯”中国研究生网络安全创新大赛，全国二等奖
- 2024，全国大学生信息安全竞赛（作品赛），全国二等奖
- 2022，中国高校计算机大赛网络技术挑战赛，全国总决赛二等奖
- 2022，全国大学生系统能力大赛操作系统设计赛, 优秀指导教师
- 2021，全国大学生信息安全竞赛（作品赛），全国二等奖

Projects

Model Stealing Attacks and Defenses for Deployed AI Systems

This research investigates the emerging security threat of AI model stealing and reverse engineering in the context of modern deployment paradigms, where trained AI models are no longer confined to cloud servers but are increasingly deployed on end devices, including mobile phones, embedded systems, edge computing nodes, and even directly within Web browsers through technologies such as WebAssembly and JavaScript-based runtimes. While this deployment trend significantly improves latency, availability, and privacy by enabling local inference, it also exposes AI models to adversaries who can directly interact with, inspect, and manipulate the execution environment. As a result, proprietary models—often representing substantial intellectual property, data curation effort, and training cost—face serious risks from model extraction, parameter reconstruction, architecture inference, and functional imitation attacks. Existing defenses at the algorithmic level are often insufficient in these settings because attackers may gain access to the deployed model artifacts rather than relying solely on black-box queries. Our research focuses on developing new model stealing attacks and a system-level defense strategy that integrates deep learning compilers, binary rewriting, and trusted execution technologies to raise the bar against both static and dynamic model stealing attacks.


Program Hardening Framework Against Low-Level Exploitation

This research topic focuses on program hardening technologies that strengthen software systems against low-level exploitation such as memory corruption attacks. Program hardening seeks to raise the attacker’s cost by fundamentally altering how programs are structured, represented, and executed, rather than relying solely on vulnerability elimination. Within this scope, program partitioning, code obfuscation, and Trusted Execution Environments (TEEs) form three mutually reinforcing pillars. Program partitioning restructures applications by separating security-critical components from less trusted logic, thereby minimizing the attack surface. Obfuscation further complicates exploitation by obscuring control flow, data layouts, and memory access patterns, degrading the attacker’s ability to reliably construct exploits or reuse gadgets even in the presence of vulnerabilities. TEEs complement these software-level defenses by providing hardware-enforced isolation, ensuring that sensitive code and data remain protected even if an attacker gains root privileges.


Attacks and Defenses in Connected Autonomous Vehicles

With the development of the automotive industry, the security of connected and autonomous vehicles (CAVs) has become a hot research field in recent years. However, previous studies mainly focus on the threats and defending mechanisms from the networking perspective, while newly emerging attacks are targeting more core components of CAVs such as OS and AI. Therefore, the defense methods against these attacks are urgently needed. In this paper, we revisit emerging attacks and their technical countermeasures for CAVs in a layered inventory, including in-vehicle systems, V2X, and self-driving. This project aims to provide insights into potential new attack vectors and their countermeasures for CAVs. We hope to shed light on future research in this area.


Securing Industrial Control Systems

Industrial control systems generally lack security design from the beginning. The design mainly considers the real-time performance, reliability and stability of the system, sacrificing security for real-time performance. Traditional IT security solutions such as identity authentication, authorization, and encryption are not suitable for industrial control systems. There are a large number of vulnerabilities in industrial devices and protocols. According to the national CNVD statistics, as of September 2021, the number of industrial control system vulnerabilities has exceeded 3,100, while security vendors, hackers and other organizations may have far more vulnerabilities than CNVD statistics. However, key industrial devices are closed and difficult to update. These loopholes may be maliciously exploited at critical moments, resulting in production theft, monitoring, and destruction. Furthermore, the accelerated integration of IT&OT has led to the exposure of more and more security threats since the industrial control systems are changing from close to open, from stand-alone to interconnected, and more and more devices are directly exposed to attackers. This project aims to secure PLC/RTU in industrial control systems. The involved techniques include hardware secure accelerator, device fingerprinting, lightweight identity authentication, etc.


Students

Copyright © 2024 by Hao Han (as of Aug. 2024)