Lu Li

img_lu.jpg
Email luli1[at]sas.upenn.edu

I am a fifth year PhD student in the Applied Math and Computational Science (AMCS) program at University of Pennsylvania, working on LLM post-training and reasoning, and long-horizon multi-turn tool-calling agents and credit assignment.

Currently, I focus on building agentic capabilities for large language models as autonomous agents at ByteDance Seed team and contributed to Seed 2.X.

Previously, I received a B.A. in Mathematics (Honors) and Computer Science from Macalester College.

News

Feb 01, 2026 Started internship at ByteDance Seed, working on LLM mid-training, post-training, and agentic RL.
Aug 01, 2025 Our work STRICT: Stress-Test of Rendering Image Containing Text has been accepted to EMNLP 2025 (main conference)!
Jun 16, 2025 Started internship at Adobe Research, working on RL for hyper-personalized LLMs.
Jan 22, 2025 Our work MAP: Low-compute Model Merging with Amortized Pareto Fronts via Quadratic Approximation has been accepted to ICLR 2025!
Aug 01, 2021 I started my PhD at UPenn.

Selected publications
* indicates equal contribution

  1. map.png
    MAP: Low-compute Model Merging with Amortized Pareto Fronts via Quadratic Approximation
    Lu Li*, Tianyu Zhang*, Zhiqi Bu*, Suyuchen Wang, Huan He, Jie Fu, Yonghui Wu, Jiang Bian, Yong Chen, and Yoshua Bengio
    In International Conference on Representation Learning , 2025
    arXiv PDF Video Code Poster
  2. strict.png
    STRICT: Stress Test of Rendering Images Containing Text
    Lu Li*, Tianyu Zhang*, Xinyu Wang*, Zhenghan Tai, Jijun Chi, Jingrui Tian, Hailin He, and Suyuchen Wang
    In EMNLP , 2025
    arXiv Code Poster
  3. vcr.png
    VCR: A Task for Pixel-Level Complex Reasoning in Vision Language Models via Restoring Occluded Text
    Tianyu Zhang*, Suyuchen Wang*, Lu Li, Ge Zhang, Perouz Taslakian, Sai Rajeswar, Jie Fu, Bang Liu, and Yoshua Bengio
    In International Conference on Representation Learning , 2025
    arXiv PDF Code Poster
  4. ai4gcc2.png
    AI for Global Climate Cooperation: Modeling Global Climate Negotiations, Agreements, and Long-Term Cooperation in RICE-N
    Tianyu Zhang*, Andrew Williams*, Phillip Wozny*, Kai-Hendrik Cohrs*, Koen Ponse, Marco Jiralerspong, Soham Rajesh Phade, Sunil Srinivasa, Lu Li, Yang Zhang, and 5 more authors
    In Forty-second International Conference on Machine Learning , 2025
    arXiv Video Code Poster Website
  5. frontiers4.png
    Minimal Cycle Representatives in Persistent Homology Using Linear Programming: An Empirical Study With User’s Guide
    Lu Li, Connor Thompson, Gregory Henselman-Petrusek, Chad Giusti, and Lori Ziegelmeier
    Frontiers in Artificial Intelligence, 2021
    arXiv HTML Video

Selected projects

  1. sprite.png
    SPRITE: debiaSed PRetraining of Transformers for Individaul Treatment Effects Estimation
    Main Contributor
    Pretrained a GPT-2 model using electronic health records (EHRs) of over 3 million patients on over 1 billion medical events.