Our lab conducts research at the intersection of security, privacy, machine learning, and human-computer interaction. Our work spans three major research directions.
AI Security
AI SecurityActive
We investigate the security and trustworthiness of machine learning models, including adversarial attacks, data poisoning, backdoor attacks on model merging, and environmental injection attacks on AI agents. Our work addresses threats across the ML pipeline — from training-time poisoning to deployment-time adversarial manipulation — and explores defenses such as automated vulnerability repair.
Related projects
Trustworthy AI agents, with Kai-Wei Chang, Cho-Jui Hsieh, and Nanyun Peng (Coefficient Giving)
Bruinweb: Amazon Nova AI Challenge — Trusted Software Agents
We study privacy risks and compliance in software systems, including GDPR enforcement, personal information disclosure in online communities, location privacy in recommendation systems, and user perceptions of data collection in IoT and public WiFi environments. Our research combines automated program analysis with empirical user studies to advance privacy protection.
Related projects
Privacy regulations and software development, for better rulemaking and compliance (NSF)
Privacy-preserving mobility data generation (NSF)
Private data analytics, synthesis, and sharing for smart-city mobility research (NSF)
Enforcing security and privacy policies to protect research data, with Kai-Wei Chang (NSF)
We analyze the security of software systems including voice-controlled platforms, IoT ecosystems, authentication protocols, smart home automations, and extended reality (XR). Our work spans vulnerability discovery in voice assistants, permission analysis, OAuth security verification, WebAssembly runtime fuzzing, and usable security tools for end users.
Related projects
CAREER: Secure voice-controlled platforms (NSF)
Safe, private, and secure home automation, with Lujo Bauer and Limin Jia at CMU (NSF)
Perception integrity in virtual reality (Meta Research Award)
Exploit generation using reinforcement learning, with Yu Feng (Google Research Scholar Award)