Hi, I’m Jingwen, a PhD student in the Columbia Core AI Lab (CAIL) and Theory of Computation Group at Columbia University, where I am extremely fortunate to be advised by Daniel Hsu and Alex Andoni. I obtained my undergraduate degree at UC San Diego, where I majored in mathematics and computer science. During my time at UCSD, I’m extremely honored to work with glorious Sanjoy Dasgupta and Russell Impagliazzo, who sparked my interest in ML theory and TCS research. I was also very lucky to work with Xiaolong Wang, who introduced me to CS research.

My recent research focuses on developing mathematical frameworks to understand and improve large language models, spanning model architecture, training/optimization, and self-improvement.

Experience

• Microsoft Research (Summer 2026), working with Akshay Krishnamurthy, Dylan Foster, Nived Rajaraman and Bingbin Liu
• Computer and Information Science at the University of Pennsylvania (Summer 2025), hosted by the amazing Surbhi Goel
• Simons Institute for the Theory of Computing, UC Berkeley (Fall 2024), Modern Paradigms in Generalization and Large Language Models and Transformers

Publications

Fixed Universal Transformers
Jingwen Liu, Alexandr Andoni, Daniel Hsu.
arXiv

Less Data, Faster Training: repeating smaller datasets speeds up learning via sampling biases
Jingwen Liu, Ezra Edelman, Surbhi Goel, Bingbin Liu.
International Conference on Machine Learning (ICML), 2026.
Contributed talk, Workshop on Scientific Methods for Understanding Deep Learning, ICLR, 2026.
arXiv | Poster

Group-realizable multi-group learning by minimizing empirical risk
($\alpha$-$\beta$) Navid Ardeshir, Samuel Deng, Daniel Hsu, Jingwen Liu.
The 37th International Conference on Algorithmic Learning Theory (ALT), 2026.
arXiv

Fast attention mechanisms: a tale of parallelism
Jingwen Liu, Hantao Yu, Clayton Sanford, Alexandr Andoni, Daniel Hsu.
Advances in Neural Information Processing Systems (NeurIPS), 2025.
arXiv | Poster

Group-wise oracle-efficient algorithms for online multi-group learning
(random order) Samuel Deng, Daniel Hsu, Jingwen Liu.
Advances in Neural Information Processing Systems (NeurIPS), 2024.
arXiv | Poster

VideoINR: Learning Video Implicit Neural Representation for Continuous Space-Time Super-Resolution
Zeyuan Chen, Yinbo Chen, Jingwen Liu, Xingqian Xu, Vidit Goel, Zhangyang Wang, Humphrey Shi, Xiaolong Wang.
Conference on Computer Vision and Pattern Recognition (CVPR), 2022.
arXiv | website

Teaching and Service

I have served as a reviewer for: COLT 2025, 2026, ICML 2025, NeurIPS 2024, 2025.

I was also a volunteer for Pre-Submission Application Review (PAR) Program which provides reviews of statements and CV from students applying to the CS PhD Program at Columbia.

I was a teaching assistant for the following classes:

• Theoretical Foundations of LLMs, by Daniel Hsu, spring 2025, Columbia University
• Machine Learning Theory, by Daniel Hsu, spring 2024, Columbia University
• Machine Learning, by Sanjoy Dasgupta, winter 2023, UCSD
• Mathematics for Algorithms and Systems, by Russell Impagliazzo, fall 2022, UCSD
• Software Project for Computer Architecture, by Hung-Wei Tseng, summer 2022, UCSD
• Introduction to Programming, by Gerald Soosairaj, spring 2022, UCSD

I was a grader for the following classes:

• Abstract Algebra, fall 2021, UCSD
• Differential Equations, spring 2021, UCSD
• Mathematical Reasoning, winter 2021, UCSD

Miscellaneous

I am deeply grateful to my advisors and many others in the community for their great mentorship, including Bingbin Liu, Surbhi Goel, Samuel Deng, Yuhao Li, Hantao Yu, Clayton Sanford, Navid Ardeshir, Samantha Chen, Geelon So, Ezra Edelman, Sihan Liu and this list goes on and on… I would like to pass on this culture, so feel free to reach out to me if you think I might be helpful!

I like playing ping pong, climbing, skiing and karaoke :)

Website template follows the tradition of Daniel’s students Geelon So and Samuel Deng.