About Me

I’m a third-year computer science PhD student at Northwestern University’s theory group. I am generally interested in theoretical computer science. I’m generally interested in theoretical aspects of machine learning.

Before beginning my PhD studies at Northwestern University, I completed both my undergraduate and master’s degrees at the same institution. (Go Cats!)

Research

Hallucination is a Consequence of Space-Optimality: A Rate-Distortion Theorem for Membership Testing
with Jingwei Li. ICML 2026 (spotlight).
arXiv version

Summary: Large language models often hallucinate with high confidence on "random facts" that lack inferable patterns. We formalize the memorization of such facts as a membership testing problem, unifying the discrete error metrics of Bloom filters with the continuous log-loss of LLMs. By analyzing this problem in the regime where facts are sparse in the universe of plausible claims, we establish a rate-distortion theorem: the optimal memory efficiency is characterized by the minimum KL divergence between score distributions on facts and non-facts. This theoretical framework provides a distinctive explanation for hallucination: even with optimal training, perfect data, and a simplified "closed world" setting, the information-theoretically optimal strategy under limited capacity is not to abstain or forget, but to assign high confidence to some non-facts, resulting in hallucination. We validate this theory empirically on synthetic data, showing that hallucinations persist as a natural consequence of lossy compression.

Agnostic Learning of Arbitrary ReLU Activation under Gaussian Marginals
with Aravindan Vijayaraghavan. COLT 2025.
arXiv version | conference version | Recorded virtual talk

Summary: We gave the first algorithm for agnostic PAC learning of an arbitrarily biased ReLU neuron under Gaussian input distributions, up to constant approximation. We also showed hardness separation bewteen SQ (statistical query) and CSQ (correlational statistical query) models for this problem. In particular, most gradient-based algorithm would fail to obtain constant approximation.

To Store or Not to Store: a graph theoretical approach for Dataset Versioning
with Jingwei Li, Pattara Sukprasert, Samir Khuller, Amol Deshpande, and Koyel Mukherjee. IPDPS 2024.
arXiv version | conference version

Summary: We study a graph-theoretic framework for dataset versioning that optimizes storage costs while maintaining retrieval costs of different versions. On the theory side, we showed the first hardness of approximation results and gave provably near-optimal algorithms for tree-like graphs (bounded treewidth). Our findings also led to better practical heuristics, providing up to 1000x speedup for the "MinSum Retrieval" problem on real-world Github repos.

Education

Northwestern University, Evanston, Illinois (2019-2023)
B.A. in Math and M.S. in Computer Science

Northwestern University, Evanston, Illinois (2023-present)
Ph.D. student in Computer Science

Awards

Ph.D. student research award (Northwestern CS department), 2024-2025.

Barris Award for outstanding TA, 2024 Fall quarter.

Junior Career Award in Mathematics (Northwestern math department), 2021-2022.

Other Experience

Ross Mathematics Program
I attended Ross Asia in 2018 as a first year student (and had a great time!), in 2019 as a junior counselor, and in 2020 as a counselor.

Directed Reading Program
In 2021 Spring, I participated in Math department’s directed reading program, where I read 6 chapters of Linear Representations of Finite Groups with Wenyuan Li.

IDEAL Summer Intern
In 2025 Summer, I interned at Toyota Technological Institute at Chicago (TTIC) and studied various topics related to transformers. I was hosted by prof. Zhiyuan Li.

Miscellaneous

I have written two articles on 知乎 (Chinese version of Quora?) about theoretical computer science. Here’s the link.

I lived in Beijing for the first 18 years of my life. What I like the most about the city is its subway system and shared bicycles, which make travelling very easy and cheap. (I’m not a big fan of driving, due to the traumatic experience of looking for parking spots for half an hour.)