"It's not that life has dreams, but that dreams create life."
「人生に夢があるのではなく、夢が人生をつくるのです」
— Taeko Uzuki
Eddy Luo (罗威迪 & 一ノ瀬 エイジ), an incoming Ph.D. student at University of Georgia, where I will be advised by Prof.Xiang Zhen. I am also fortunate to be co-advised by Prof. Chaowei Xiao at the University of Wisconsin–Madison, a mentor I deeply respect and am sincerely grateful to. Previously, I served as a research assistant at the OSU NLP Group and the ICICLE Institute advised by Prof. Yu Su.
Eddy warmly welcomes collaboration opportunities and supports undergraduates who want to apply for a PhD program. He hopes we can conduct significant research together. Please feel free to contact him at Email: luo.1455[shift+2]uga[dot]edu. どうぞよろしくお願いします!
Using interpretability methods, discover security vulnerabilities in AI systems, including foundation models and AI agents, and develop corresponding defense and detection algorithms, including safety alignment strategies.
Leverage AI to drive defense and attack strategies on systems, including web system and operating system.
Develop lifelong learning AI frameworks and defense systems by utilizing reinforcement learning, cognitive science, bio-inspired algorithms, active learning, and so on.
Two of our works, AGrail: A Lifelong Agent Guardrail with Effective and Adaptive Safety Detection and Disentangling Memory and Reasoning Ability in Large Language Models have been accepted by ACL'2025 main conference. Thanks to my collaborators.
Our work JailBreakV-28K: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks wins $20,000 SafeBench Prize for Advancing MultiModal Large Language Model Security Benchmarking from Center for AI Safety.
I will join the University of Georgia as a PhD student in August 2025.
Doxing via the Lens: Revealing Location-related Privacy Leakage on Multi-modal Large Reasoning Models
AGrail: A Lifelong Agent Guardrail with Effective and Adaptive Safety Detection
JailBreakV-28K: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks
