PDF

Profile

Undergraduate at Fudan University (School of Mathematical Sciences), pursuing a double degree in Information and Computing Science and Artificial Intelligence.

GPA: 3.96/4.00 (Fall 2025), 4.00/4.00 (Spring 2026); Cumulative: 3.98/4.00; Rank: 3/213 in cohort.

Education

Fudan University, Shanghai, China — Sep. 2025 – Present

B.S. Candidate in Information and Computing Science & Artificial Intelligence (Double Degree), School of Mathematical Sciences.

Research Interests

Optimization, control, and reinforcement learning for sequential decision-making.

Mathematical modeling for real-world problems.

Selected Open-Source Projects

token-verification-miragePaper · AI4Math Workshop

  • Poster, ICML 2026 Workshop on AI for Math (AI4Math); solo-authored, full pipeline (dataset selection, generation, evaluation design, analysis, writing)
  • Evaluation-protocol choices alone (pooling, in-sample scoring, direction-agnostic AUROC) shift apparent verification AUROC by up to 0.18
  • Under corrected within-problem, leave-one-run-out, fixed-direction scoring, shallow token statistics cluster at only 0.60–0.75 AUROC; final-token entropy drops from 0.72–0.75 to 0.47–0.48 once the direction-agnostic reporting artifact is removed

code-not-textDemo

  • Cross-domain measurement study across 31,040 math, science, and coding runs: AoA 0.958 (math), 0.799 (science), 0.434 (coding, below the random-direction baseline)
  • Rules out five alternative explanations for the coding failure (model capacity, label scarcity, surface noise, feature coverage, judge framing) — all converge to the same ceiling; frames the result as measurement non-invariance

TinyLoRA-GRPO-CoderDeepWiki

  • Adapts Learning to Reason in 13 Parameters (Morris et al., 2026) to competitive programming: trains 32 shared scalars via GRPO on Qwen2.5-Coder-3B, rewarded by real g++ compile-and-run outcomes

microgpt.cpp

  • Minimal GPT (autograd, multi-head attention, Adam) from scratch in ~300 lines of C++, inspired by Karpathy’s teaching gist

Academic Service

Reviewer, ICML 2026 Workshop on AI for Math (AI4Math), 2026

Talk: Reinforcement Learning: From Bandits to PPOApr 18, 2026 (PDF notes)

  • Overview covering multi-armed bandits, MDPs, policy gradient, and PPO

Skills

Programming: Python, C++, C · ML/AI: PyTorch, ML experimentation, LLM evaluation, RL basics · Tools: Git, GitHub, Linux, LaTeX, Markdown · Language: Chinese(native) English(fluent)

Selected Course Grades

SemesterCourseGrade
Fall 2025ProgrammingA
 Analytic GeometryA
 Mathematical Analysis IA
 Advanced Algebra IA-
Spring 2026Mathematical Analysis IIA+
 Advanced Algebra IIA
 Foundations of Software for AIA
 Introduction to AIA

Community Involvement