CV
Chenqing Zhu
Ph.D. Student, Southeast University
Research Interests: LLM Security, Multimodal LLM Security, Federated Learning, Backdoor Attacks, AI Safety Evaluation
Email: chenqingzhu [at] seu [dot] edu [dot] cn / chenqing_zhu [at] outlook [dot] com
Homepage: https://chenqing-zhu.github.io/
Education
Southeast University - Ph.D. Student
Jiangsu, China
09/2024 - Present
Major: Cyberspace Security (Artificial Intelligence Track)
Supervisor: Prof. Songze Li
Hong Kong University of Science and Technology (Guangzhou) - Master
Guangdong, China
09/2022 - 06/2024
Major: Internet of Things
Supervisors: Prof. Songze Li and Prof. Danny Hin Kwok Tsang
Soochow University - Bachelor
Jiangsu, China
09/2018 - 06/2022
Major: Software Engineering, GPA: 3.9/4.0, Ranking: 1/79
Selected Publications
C. Zhu, Y. Dai, Y. Tian, Q. Li, and S. Li, “When the Aggregator Cheats: Data-Free Backdoors in Federated LLM-based QA Systems,” in Proceedings of the 35th USENIX Security Symposium, 2026. Accepted.
C. Zhu and S. Li, “Client-Driven Federated Learning under Dynamic Mixtures of Distributions,” in Proceedings of the 21st International Conference on Wireless Algorithms, Systems, and Applications (WASA), 2026. Accepted.
J. Xie, C. Zhu, and S. Li, “FedMeS: Personalized Federated Continual Learning Leveraging Local Memory,” Federated Learning Workshop at IJCAI, 2023; arXiv preprint arXiv:2404.12710, 2024.
Research Experience
Data-Free Backdoors in Federated LLM-based QA Systems
Research Project | Southeast University | 2024 - 2026
- Proposed a data-free server-side backdoor attack against federated LLM-based QA systems, where a malicious aggregator implants advertisement-style backdoors without accessing client raw data.
- Developed a gradient-inversion-based pipeline to reconstruct domain-relevant semantic cues and synthesize poisoned pseudo-QA data for deployment-time backdoor injection.
- Evaluated the attack across medical, mental health, and legal QA scenarios under multiple LLMs and Full FT/LoRA settings; accepted by USENIX Security 2026.
Client-Driven Federated Learning under Dynamic Mixtures of Distributions
Research Project | HKUST(GZ) / Southeast University | 2023 - 2026
- Proposed a client-driven federated learning framework for dynamic mixtures of client distributions with asynchronous client-initiated model updates.
- Designed a server-side cluster repository to support personalized model adaptation under changing client data distributions.
- Evaluated the method on rotated FashionMNIST, CIFAR-100, MiniImageNet-100, and digit-domain benchmarks; accepted by WASA 2026.
FedMeS: Personalized Federated Continual Learning
Collaborative Research Project | HKUST(GZ) | 2022 - 2024
- Collaborated on FedMeS, a personalized federated continual learning framework that leverages local memory to mitigate client drift and catastrophic forgetting.
- Contributed to algorithm design and theoretical analysis, including memory-assisted gradient calibration during training and personalized inference with local-memory-based KNN Gaussian modeling.
- Evaluated the method across continual federated benchmarks with varying datasets, task distributions, and client numbers; accepted by FL-IJCAI’23 and released as an arXiv preprint.
Industry Collaboration
OmniTrust: Enterprise LLM Safety and Data Security Evaluation
Project Organizer / Client-facing Lead | Southeast University | 2025 - Present
- Helped build OmniTrust, an enterprise-oriented LLM safety evaluation framework for assessing data security, compliance risks, and unsafe model behaviors in real-world deployment scenarios.
- Organized and led project execution, including evaluation scope definition, test case design, client communication, task coordination, and assessment report delivery.
- Designed evaluation dimensions covering privacy leakage, enterprise-sensitive information exposure, prompt injection, harmful outputs, data boundary violations, and China AI regulatory compliance; led a case study for a multinational chemical company in China.
Internships
Back-end Development Intern, ByteDance
Hangzhou | 03/2022 - 08/2022
- Worked on Linux kernel and virtualization-related backend testing.
- Studied mainstream testing and development frameworks for Linux kernel and CPU/GPU virtualization.
- Participated in the development of FAST, an automated health-status check system for backend network servers.
Azure Network Support Engineer Intern, Microsoft
Wuxi | 07/2021 - 09/2021
- Worked on Azure networking, including virtual networks, load balancers, VPN, and IaaS network connectivity.
- Assisted in troubleshooting customer-side network connectivity and performance issues.
- Gained experience in interactions between on-premise devices and cloud computing centers.
Teaching Experience
Teaching Assistant: AI Security and Privacy, Southeast University
03/2025 - 06/2025
- Assisted in organizing a seminar-style course on AI security and privacy.
- Supported course logistics, project organization, and student evaluation.
- Covered topics related to trustworthy AI, privacy, security, and LLM safety.
Technical Skills
Programming & Deep Learning
- Python, PyTorch, Hugging Face Transformers, experiment automation
- Java, SQL, GoLang, etc.
Systems & Tools
- Linux, Git, Docker, LaTeX
AI Security & LLMs
- LLM fine-tuning, LoRA/PEFT, federated LLM training, gradient inversion, backdoor attacks, prompt injection evaluation
Languages
- Chinese: Native
- English: Professional working proficiency
- TOEFL: 105
- GRE: 324 + 3.5