Zhao (Bill) Xu

Room 386

Engineering VI, UCLA

Los Angeles, CA 90095

As a new beginning, I am maintaining my new homepage – still under construction.

I am a Ph.D. student at UCLA, and fortunate to be supervised by Prof. Cho-Jui Hsieh.

I received my M.Phil. degree from The Hong Kong University of Science and Technology, supervised by Prof. Hao Liu and Prof. Hui Xiong. Prior to my M.Phil., I received a Bachelor’s degree from the Department of Electronic Engineering at Tsinghua University, where I was fortunate to be advised by Prof. Yong LI.

My primary research focus is on building and enhancing trustworthy foundation models (e.g., LLMs, VLMs). More specifically,

(i) how to better develop FMs that maintain robust performance under varying conditions while ensuring harmless and reliable responses;
(ii) how to further boost the trustworthy performance of FMs without increasing model sizes or pre-training dataset sizes, particularly as the benefits from scaling laws begin to plateau.

I am fortunate to have conducted some initial research in the areas of safety, robustness, and truthfulness under the guidance of all my research advisors alongside many wonderful collaborators, and I am passionate to continue studying these problems in my Ph.D.

News

Sep 22, 2025	I’ve taken on my PhD commitment.
Aug 05, 2025	We are awarded the Audience Appreciation Award (only 2 out of all accepted papers) at KDD 2025! Thanks a lot for all the support from my advisors and collaborators!
May 30, 2025	I am honored to be recognized as an Outstanding Reviewer (Top 10%) for KDD 2025! I am grateful for the opportunity to contribute to the research community.
Apr 13, 2025	🎉🎉 Our work, LLMLight, received the 🏆 Special Grand Award & 🏆 Golden Medal (the highest award!!) at the 50th International Exhibition of Inventions Geneva. Many thanks to Siqi for his continuous dedication and to our advisors for their huge support!

Work Calendar

Selected publications

NeurIPS’25

Bag of tricks: Benchmarking of jailbreak attacks on llms

Zhao Xu, Fan Liu, and Hao Liu

Advances in Neural Information Processing Systems, 2024

Bib PDF Code 🤗 Model

@article{xu2024bag,
  title = {Bag of tricks: Benchmarking of jailbreak attacks on llms},
  author = {Xu, Zhao and Liu, Fan and Liu, Hao},
  journal = {Advances in Neural Information Processing Systems},
  volume = {37},
  pages = {32219--32250},
  year = {2024},
  model = {https://huggingface.co/usail-hkust/JailJudge-guard}
}

KDD’25

Llmlight: Large language models as traffic signal control agents

Siqi Lai, Zhao Xu, Weijia Zhang, Hao Liu, and Hui Xiong

In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 1, 2025

Bib PDF Code 🌐 Website 🤗 Model

@inproceedings{lai2025llmlight,
  title = {Llmlight: Large language models as traffic signal control agents},
  author = {Lai, Siqi and Xu, Zhao and Zhang, Weijia and Liu, Hao and Xiong, Hui},
  booktitle = {Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 1},
  pages = {2335--2346},
  year = {2025},
  model = {https://huggingface.co/collections/usail-hkust/lightgpt-673ac5a619cbbe309165b56d}
}