Hi, I am Ziyu. I am a Ph.D. student in Electrical and Computer Engineering at Binghamton University. My research focuses on efficient and trustworthy AI systems, particularly the security and protection of large language models (LLMs).
I study how to protect AI models from extraction and enable secure, efficient on-device inference. My work brings together model-level defenses and trusted execution environments (TEEs), with attention to the balance between security, model utility, and system performance.
Previously, I received my M.S. in Electrical Engineering from Northeastern University in 2022 and my B.S. in Electrical Engineering from Pennsylvania State University in 2020.
61st ACM/IEEE Design Automation Conference.
A two-branch architecture that protects AI models in a TEE while reducing secure memory use and inference overhead.
ACM Transactions on Design Automation of Electronic Systems, 2024. Published online December 2024; Vol. 30(1), Article 13.
An experimental study of fault-injection vulnerabilities in AXI and PCIe communication, with protocol-specific defenses.
IEEE/ACM International Conference on Computer-Aided Design.
A lightweight secure branch that protects on-device AI models with minimal inference latency overhead.
Xin Chen, Qiyan Zhang, Ziyu Liu, Yifei Sun, Q. M. Zhang
Applied Physics Letters, 120(16), 162902.
Qiyan Zhang, Xin Chen, Bing Zhang, Tian Zhang, Wengchang Lu, Zhe Chen, Ziyu Liu, Seong H. Kim, Brian Donovan, Ronald J. Warzoha, Enrique D. Gomez, J. Bernholc, Q. M. Zhang
Matter, 4(7), 2448–2459.