Kelvin Yuxiang Huang

Undergraduate Researcher | CS & Stats @ UC Berkeley

portrait.jpeg

yxkelvinhuang[at]berkeley.edu

I am a senior undergraduate student at the University of California, Berkeley, double majoring in Computer Science and Statistics.

My research interests lie broadly in Artificial Intelligence, with a specific focus on Trustworthy AI, Vision-Language Models (VLMs), and Agentic Systems. My current goal is to build AI systems that are robust against adversarial attacks and transparent through structured observability.

Currently, I am a Researcher on Trustworthy Agentic AI at UC Berkeley, advised by Prof. Dawn Song, focusing on structured observability for agent systems. Simultaneously, I conduct independent research on adversarial robustness for LVLMs at the Blender Lab (UIUC), advised by Prof. Qingyun Wang (now at William & Mary).

Previously, I was selected for the Summer Research Programme (SRP) at HKU-EEE, where I developed a physics-aware 3D generation pipeline advised by Prof. Xihui Liu. I also served as a Lead Developer at the Levi Lab (UC Berkeley), engineering computational frameworks for vision and eye movement modeling under Prof. Dennis Levi.

I am actively applying for Ph.D. programs for Fall 2026.

news

Nov 19, 2025 Thrilled to share that our paper on Trustworthy Agentic AI has been accepted to the AAAI 2026 TrustAgent Workshop! 🎉
Nov 05, 2025 Thrilled to share that our paper on VLM adversarial robustness has been accepted to the AAAI 2026 AIGOV Workshop! 🎉
Sep 21, 2025 Excited to start as a researcher advised by Prof. Dawn Song to work on Trustworthy Agentic AI.
Apr 14, 2025 Excited to kick off my research journey at Blender Lab (UIUC)! I will be working on VLM Robustness, advised by Prof. Qingyun Wang.
Mar 12, 2024 Honored to be selected for the Summer Research Programme (SRP) at HKU, advised by Prof. Xihui Liu.

selected publications

  1. AAAIw
    ATLAS: Shielding Geolocation in LVLMs with Zero-Query Universal Perturbations
    Kelvin Yuxiang Huang, Yi R. Fung, Yue Xiao, Daniel Runfola, and Qingyun Wang
    In AAAI-26 Workshop on AI for Governance (AIGOV), 2026
    Accepted (Poster, Non-archival)
  2. AAAIw
    AgentTrace: A Structured Logging Framework for Agent System Observability
    Adam ALSayyad*, Kelvin Yuxiang Huang*, and Richik Pal*
    In AAAI-26 Workshop on AI for TrustAgent, 2026
    Accepted (Poster, Non-archival)
    * Equal contribution