About me
I pursue a central question: how can foundation models meet practical constraints without compromising capability or trust? I treat efficiency, safety, and multimodal learning as connected system-design problems rather than isolated benchmarks, and I work across the full research cycle — translating a concrete challenge into a precise learning objective, designing and implementing the method, and building the rigorous evaluation pipeline that tests whether it holds up.
That question has taken my work in three directions that reinforce one another: making large language models efficient enough to run under real memory constraints, making them robust against adversarial misuse, and extending learning across language and vision. Each thread has resulted in peer-reviewed publications, most recently a co-first-authored paper in the Findings of ACL 2026.
I hold a Ph.D. in Computer Science from UC Irvine, where my research on trustworthy and efficient foundation models has accumulated 360+ citations. I have supported graduate courses in machine learning and deep learning at UCI, and I contribute to the research community as a reviewer for peer-reviewed journals.
Areas of Expertise
-
Efficient Foundation Models
Low-bit quantization that makes large language models practical under tight memory budgets.
-
Multimodal Learning
CLIP and image–text representation learning, including multimodal classification and evaluation pipelines.
-
Trustworthy AI & Safety
Adversarial training and safety classification for LLMs, hardening models against gradient-optimized jailbreaks.
-
Programming & Frameworks
PyTorch, Hugging Face Transformers, scikit-learn; Python, SQL, MATLAB, C/C++, Java.