Hi there, I'm Zexin Xu (徐泽鑫)! 👋

💡My name is pronounced as "Zeh-Shin She", and I also go by Jason.

I am a second-year Ph.D. student in Computer Science at the University of Texas at Dallas, co-advised by Dr. Yapeng Tian and Dr. Wei Yang. My current research focuses on egocentric multimodal agents and multimodal human state understanding, with additional interests in post-training and learning technology. Recently, I led team ASTRO in the Amazon Trust AI Challenge, where we were selected as one of the five Red Teams globally and made it into the finals.

Prior to UTD, I completed both my M.S. and B.S. in Computer Science and Engineering at The Ohio State University, where I worked with Dr. Yu Su. During my master's program, I led the software engineering team for SalsaBot in the Amazon Alexa Prize SimBot Challenge. In the field of learning technology, I collaborated with Dr. Di Xu (UCI), Dr. Qiujie Li (NTU), and Dr. Zilu Jiang (JHU) on TimeWise, an educational conversational agent project. Our team participated in the Tools Competition 2024 and was selected as one of the five finalists among approximately 2,000 submissions.

2026

Joined Dr. Yapeng Tian's research group at UT Dallas. Really excited for this new research journey!

Mar24
2025

Crossed the 100-citation milestone! I’m deeply grateful to my collaborators and mentors for making this possible. Thank you!

Nov06

Our Team ASTRO has been selected as one of the 4 finalists in the Amazon Trust AI Challenge! Check this out!

Jun24

Our paper "SoK: Efficiency Robustness of Dynamic Deep Learning Systems" accepted to USENIX Security '25! Check our LinkedIn Post!

Jun15

Grateful to my amazing coauthors and advisors! Our LLM4SR (LLM for scientific research) survey paper is now on arXiv!

Jan08
2024

Led team ASTRO to be selected as one of the five Red Teaming groups for the Amazon Trust AI Challenge. See you in Seattle! 🚀

Nov01

I'm so blessed to be married to the love of my life, Jessie! ❤️ Here's to forever together 🥂

Oct23

Our project TimeWise has been selected as one of the five finalists in the Tools Competition 2024! 🎉

Apr16
2025
Efficiency Robustness of Dynamic Deep Learning SystemsUSENIX Security
Ravishka Rathnasuriya, Zexin Xu, Tingxi Li, Zihe Song, Mirazul Haque, Simin Chen, Wei Yang
@USENIX Security '25
LLM4SR: A Survey on Large Language Models for Scientific ResearcharXiv
Ziming Luo, Zonglin Yang, Zexin Xu, Wei Yang, Xinya Du
@arXiv preprint
2023
Towards a robust and generalizable embodied agentCVPR
Chan Hee Song, Jiaman Wu, Ju-Seung Byun, Zexin Xu, Vardaan Pahuja, Goonmeet Bajaj, Samuel Stevens, Ziru Chen, Yu Su
@Embodied AI Workshop at CVPR
Exploring the Role of Artificial Intelligence in Facilitating Assessment of Writing Performance in Second Language LearningLanguages
Zilu Jiang, Zexin Xu, Zilong Pan, Jingwen He, Kui Xie
@Languages
Embodied AgentPython

SalsaBot is an innovative embodied agent, specifically developed for the first Alexa Prize Simbot Challenge. The agent’s core function is to guide users through a game in a virtual setting, focusing on a seamless, user-centric experience.

ChatbotReactFlask

An LLM-based conversational agent using cognitive behavioral therapy to transform users' time management skills.

Tele-Tube

2022
ReactChrome Extension

Chrome Extension for Time-Stamping Youtube Captions