Shilun (Allan) Li

PhD Student

I am a fourth-year Ph.D. student in Mathematics at UC Berkeley, advised by Prof. Venkatesan Guruswami. Previously, I graduated from Stanford University with a Bachelor’s degree in Mathematics with honors and a concurrent Master’s degree in Computer Science (AI track). My research interests lie broadly in Coding Theory, Complexity Theory, Automated Theorem Proving, and Machine Learning.

PDF
Email
Website
https://shilun-allan-li.github.io/
Google Scholar
Shilun Li
GitHub
Shilun-Allan-Li

Experience

Machine Learning Intern at Apple
–  ·  Cupertino, CA

Highlights

  • Developed end-to-end ML systems to solve complex modeling problems.

Advisor/Lecturer at XAcademy
Jul – Aug 2023 & Jul – Aug 2025  ·  Shanghai, China

Highlights

  • Taught a 45-lecture ML/DL course (2023) to 40 students, covering neural networks, RNNs, CNNs, NLP, Transformers, and LLMs.
  • Taught a 50-lecture Multimodal Learning course (2025) to 30 students, covering joint embeddings, alignment, multimodal Transformers, and vision–language applications.

Research Collaborator at Microsoft Research
–  ·  Mountain View, CA

Highlights

  • Built a post-training framework combining MCTS, PPO, and reward models for automated theorem proving.
  • Improved Qwen2.5-7B-Instruct by 10–25% on the MATH benchmark.

Co-founder at FutureX (Futurexyz Co.)
–  ·  Palo Alto, CA

Highlights

  • Built a Web3 platform for professional networking and events.
  • Raised $300K in seed funding from MiraclePlus (formerly Y Combinator China).

AI Developer Intern at Tencent
–  ·  Shenzhen, China

Highlights

  • Designed a procedural 3D city-generation algorithm in Unreal Engine 4 (C++).
  • Automated generation of roads, buildings, rivers, and parks for game environments.

–  ·  Stanford, CA

Highlights

  • Implemented Social-GAN pedestrian prediction on the JRDB dataset.
  • Evaluated predictions in real-world robot perception pipelines.

–  ·  Stanford, CA

Highlights

  • Designed a probing framework for BERT to analyze syntactic and semantic signals across intermediate layers.

Education

–

Ph.D. in Mathematics
  from UC Berkeley with GPA of 4.0/4.0

Courses

–

M.S. in Computer Science (AI track)
  from Stanford University with GPA of 4.0/4.0

–

B.S. in Mathematics with Honors
  from Stanford University with GPA of 4.0/4.0

Courses

Publications

Density Frankl–Rödl on the Sphere

Venkatesan Guruswami*, Shilun Li* · RANDOM 2025

Generalizes the Frankl–Rödl theorem to the sphere by proving a density statement for spherical avoidance problems; further establishes that all simplices are sphere Ramsey.

A Deterministic Construction of a Large Distance Code from the Wozencraft Ensemble

Venkatesan Guruswami*, Shilun Li* · IEEE Transactions on Information Theory · Presented at RANDOM 2023

Applies Sidon sets to code construction (the first such application) and presents the first explicit construction of rate-1/2 Wozencraft codes with Ω(√k) minimum distance.

Dynamics and Probability in the Toss of a Coin with Symmetric Inhomogeneous Density

Shilun Li · Preprint

Proves that coin tosses are biased — coins favor the starting face — and extends Diaconis's probability formula to general biased coins.

Awards

McBeth Family Fellowship in Mathematics

UC Berkeley — 2025

Departmental Honors in Mathematics

Stanford University — 2022

(Awarded to top 15% of graduating cohort.)

Top Three Projects

Stanford BIOME Biohackathon — 2018

AMS Special Award

Intel ISEF — 2017

(The world's largest pre-college STEM competition.)

National Best High School Student

All-China Students' Federation — 2017

(Top 10 out of 170,000 nationwide.)

First Prize (Two-time)

Provincial Science & Technology Innovation Contest — 2016

(Top 5 out of 400 in the province.)

Gold Award in Science & Gold Award in Math

S.-T. Yau Science Award — 2016

(1st out of 5,800+ global participants.)

Skills

Mathematics
Level: 3
Keywords:
  • Coding Theory
  • Complexity Theory
  • Combinatorics
  • Probability Theory
Machine Learning
Level: 3
Keywords:
  • Pytorch
  • LLM
  • GNN
  • Post Training
  • Reinforcement Learning
Python
Level: 2.5
Keywords:
  • Scikit-learn
  • NumPy
  • SQL
  • jekyll
C++
Level: 2.5
Keywords:
  • High-Performance Computing
  • GPU Programming
  • Unreal Engine 4
Theorem Formalization
Level: 2
Keywords:
  • Lean 4
  • Autoformalization
Other Languages
Level: 2
Keywords:
  • Java
  • MATLAB
  • Scala
  • R
  • Shell

Languages

English
Fluency: Native Speaker
Chinese Mandarin
Fluency: Native Speaker
Cantonese
Fluency: Elementary Proficiency

Interests

Sports
Keywords:
  • Professional Freediver
    (National Rank #45 CWB, #31 CNF)
  • Skiing
  • Tennis
  • Golf
  • Mountain Biking
Other
Keywords:
  • Photography
  • Pastry Arts
  • Bartending
  • Singing
  • Gaming