Lyuxin David Zhang

I'm an undergraduate and accelerated master's student at the University of Pennsylvania, studying Computer Science & Mathematics (B.A.) and Data Science (M.S.E.), graduating in May 2027. I'm looking for PhD opportunities starting in Fall 2027!

I’m interested in principled approaches to machine learning, combining mathematical insights with empirical research to understand how language models learn and generalize. I also think about how to make them more useful in practice, with a focus on reasoning. At UPenn, I am fortunate to work with Surbhi Goel, Eric Wong, and Dan Roth, and have collaborated with Chris Callison-Burch.

Outside of research, I play soccer and follow Shanghai Port.

Email  /  Scholar  /  Github  /  X

Lyuxin David Zhang

Research

My research interests include:

  • Looped transformers and their generalization
  • Data selection for LLM post-training
  • Multi-hop question answering and exhaustive search with LLM agents

* denotes equal contribution.
LESSER pipeline and feature-extraction FLOP reduction LESSER: Post-Training Data Selection with Output-Layer Gradients
Lyuxin David Zhang, Eric Wong, Surbhi Goel, Anton Xue
Under review at ICLR 2027
Preliminary version, Workshop on Methods and Opportunities at Small Scale @ COLM, 2026
arXiv  /  pdf  /  code
An Exhaustive-Search Benchmark for AI Agents
Lyuxin David Zhang, Hongyi James Cai, Tomer Wolfson, Dan Roth
In Progress
Decomposed Prompting Does Not Fix Knowledge Gaps, But Helps Models Say “I Don't Know”
Dhruv Madhwal*, Lyuxin David Zhang*, Dan Roth, Tomer Wolfson, Vivek Gupta
Findings of ACL, 2026
arXiv  /  pdf  /  code
DoGEN: Domain-Specific Text Detection with Parameter-Efficient Routing and Large-Ensemble Baselines
Arihant Tripathi*, Liam Dugan*, Charis Gao, Maggie Huan, Emma Jin, Peter Zhang, Lyuxin David Zhang, Julia Zhao, Chris Callison-Burch
arXiv preprint, 2025
arXiv  /  pdf

Fellowships & Awards
2026 College of Arts & Sciences Travel Grant, UPenn Center for Undergraduate Research and Fellowships ($1,500).
2026 Ruth Marcus Kanter College Alumni Society Undergraduate Research Grant, UPenn Center for Undergraduate Research and Fellowships ($1,000).
2026 Modal Academics Computing Grant, Modal Labs ($2,000).
2026 Tinker Research Grant, Thinking Machines Lab ($5,000).
2024 Pincus–Magaziner Family Undergraduate Research Fund, UPenn Center for Undergraduate Research and Fellowships ($1,000).
2022 USACO Gold Division.
2021 AIME Qualifier.

Selected Projects
Drone Racing Policy Learning with Reinforcement Learning
  • Final project for ESE 6510 (Physical Intelligence: Science and Systems); placed third in the final class challenge on deploying a simulation-trained policy on a real drone.
  • Reproduced a drone racing control method in NVIDIA Isaac Lab with PPO.
  • Tuned observations, reward shaping, and domain randomization for high-speed, robust policy learning.
Training loss curves for transformer in-context learning Training Dynamics of Transformers for In-Context Learning
  • Built a PyTorch benchmarking suite for transformer in-context linear regression.
  • Compared transformer attention dynamics with OLS and Ridge baselines.
Underflow system architecture: CLI, FastAPI backend, MySQL, and Next.js dashboard Repository Intelligence CLI
  • Built a FastAPI/Uvicorn and MySQL backend with a CLI client using the GitHub and OpenAI APIs.
  • Scanned repositories and recommended performance and cost improvements.
  • Winner of Best Use of DeFANG at PennApps 2024.

Teaching

Teaching Assistant at the University of Pennsylvania for CIS 5200 (Machine Learning), ESE 5460 (Principles of Deep Learning), and CIS 2450 (Big Data Analytics). Tutor for MATH 3120 (Linear Algebra).


Template based on Jon Barron's website.