About Me
I am a PhD student in Computer Science at the University of Southern California (USC), where I started in August 2026. I received my master’s and bachelor’s degrees in Computer Science and Technology from the University of Science and Technology of China (USTC). During my master’s studies, I was advised by Prof. Pan Peng. I am also grateful for the guidance of Prof. Christian Sohler and Prof. Lili Qiu.
My research spans algorithms and theoretical computer science, including sublinear algorithms, and AI agent evaluation, with a focus on long-term memory and tool use. I enjoy both proof-based research and hands-on experimentation. My previous work also includes satellite network optimization and high-performance computing.
You can find my CV here.
On a more personal note, I enjoy birdwatching and writing poems.
Publications
Click a title for an overview of the work.
Conference Papers
VehicleMemBench: An Executable Benchmark for Multi-User Long-Term Memory in In-Vehicle Agents
COLM 2026
Overview | Paper | Poster | Code | Dataset
Authors: Yuhao Chen, Yi Xu, Xinyun Ding, Xiang Fang, Shuochen Liu, Luxi Lin, Qingyu Zhang, Ya Li, Quan Liu, Tong Xu
An executable benchmark that evaluates whether agents can use evolving, user-specific memories to perform correct tool actions. Task success is assessed through the final state of a vehicle simulator.
Sublinear Algorithms for Estimating Single-Linkage Clustering Costs
ESA 2026
Overview (1 min) | Paper | Slides
Authors: Pan Peng, Christian Sohler, Yi Xu (alphabetical order)
We proposed new cost functions about SLC, which reflects properties of datasets, and estimated our costs in sublinear time. We studied this problem in both distance and similarity measurements, and conducted extensive experiments to validate our results. We also proved the lower bounds, and our algorithms nearly match them.
Joint Optimization of Handoff and Video Rate in LEO Satellite Networks
IEEE INFOCOM 2026
Overview (1 min) | Paper
Authors: Kyoungjun Park, Zhiyuan He, Cheng Luo, Yi Xu, Lili Qiu, Changhan Ge, Muhammad Muaz, Yuqing Yang
We propose and evaluate new algorithms for video-aware mobility management in satellite networks, jointly optimizing satellite handoff and video bitrate to improve streaming experience for both single and multiple users.
Manuscripts
Exact Direct Sum Theorems for Shallow Circuits
Authors: Xinyu Mao, Yi Xu, Guangxu Yang (alphabetical order)
We study exact direct-sum theorems for shallow circuits, including depth-3 AC⁰ and depth-4 NC⁰ circuits.
Replica Server Placement in a Satellite Network
Overview (1 min) | Paper | Slides
Authors: Zhiyuan He, Yi Xu, Cheng Luo, Lili Qiu, Yuqing Yang
Satellites can be applied as content replica servers to improve client experience in the remote locations. We optimized the placement of servers in satellite networks to reduce transmission and storage costs while considering satellite movement.
Projects
Click a title for project details.
ASC22-23 International Supercomputing Competition
Overview (1 min) | Team leader. Jan 2022 ~ May 2023
Led a team to optimize various applications, including numerical simulation and large language models (LLMs). Awarded 2nd Place among 300 teams worldwide.
VSCode-Azure DevOps Integration Extension
Overview (1 min) | Team leader. Sep 2022 ~ Dec 2022
Led the design and development of a Visual Studio Code extension enabling pull request (PR) workflows with Azure DevOps.
ISC21 International Supercomputing Competition
Overview (1 min) | Participant. Oct 2020 ~ Jun 2021
Enhanced a Fortran-based weather forecasting model (WRF) by optimizing compilation flags and parallelization strategies. Ranked 1st on the WRF application and 6th overall among 13 finalist teams in the competition.
RoboGame2020 Competition
Overview (1 min) | Participant. Jun 2020 ~ Oct 2020
Collaborated in designing and programming a robot with STM32 embedded systems for moving simulated patients to designated beds; programmed and controlled the robotic arm for task execution. Achieved Fourth Place in the second round of the competition.
Pixel-Style Graphic PC Game
Overview (1 min) | Participant. Oct 2019 ~ Jun 2020
Utilized C++ to implement features for character navigation, combat, item selection, map loading and NPC iteraction.
Talks
- Sublinear Algorithms for Estimating Single-Linkage Clustering Costs
Women in Theory 2025 Rump Session at UC-Berkeley, Jun 2025. Slides.
Student Theory Day 2025 at USTC, Mar 2025. Slides. - Experiences from an Award-Winning Team in the ASC Supercomputing Competition ASC24 Supercomputing Competition Startup at Beijing, Jan 2024. Slides.
Teaching Assistant
- Algorithms for Big Data. Spring 2025. Syllabus.
- Design and Analysis of Algorithms. Fall 2024. Syllabus.
Academic Experience
- iFLYTEK, COG Team, Core R&D Platform — Algorithm Research Intern
Hefei, China · Dec 2025 – Jul 2026. Worked on a multi-user benchmark for long-term memory and decision-making in vehicle agents. Contributed to an LLM-based pipeline that generates instruction–response pairs and filters them with programmatic validators for instruction-following training. - Microsoft Research Asia, Wireless Group — Research Intern
Shanghai, China · Aug 2022 – Mar 2023. Investigated adaptive bitrate selection and replica server placement in satellite networks; implemented signal processing on USRP and analyzed wireless sensing data in Python. - Women in Theory Workshop, Simons Institute, UC Berkeley — Participant
Jun 2025. Presented our work on sublinear algorithms for estimating single-linkage clustering costs in the student rump session. - TCS Workshop, Nanjing University — Participant
Jul 2023 and Jul 2024.
Awards and Scholarships
- First-Class Academic Scholarship, USTC (2023, 2024).
- Silver Prize, ASC22-23 International Supercomputing Competition (2023).
- Outstanding Graduate, USTC, Undergraduate (2023).
- Star of Tomorrow, Microsoft Research Asia (2022).
- Silver Award, Outstanding Student Scholarship, USTC (2019, 2020).