Project Portfolio

Project Portfolio项目作品集

Course Reports, Challenges, and Independent Builds课程报告、挑战赛与独立构建

This page is a record of serious coursework, challenge work, teaching-assistant project design, and independent building. These are not publication-style research claims yet; they are project reports that show how I formulate questions, build systems, run experiments, and explain what I learned.这个页面记录的是严肃的课程作业、挑战赛工作、助教项目设计和独立构建。它们还不是发表式科研宣称，而是展示我如何提出问题、搭系统、做实验，并解释自己学到了什么的项目报告。

Project Reports项目报告

The common thread is training: coursework made me practice mathematical modeling and controlled experiments; challenges forced validation discipline; independent builds taught me how interfaces and agents behave when ideas meet users.共同线索是训练：课程作业让我练习数学建模和受控实验；挑战赛迫使我重视验证纪律；独立构建则让我看到界面和智能体在真正遇到用户时会怎样表现。

DLfinal

Final Project / Mathematical Introduction to Machine Learning机器学习数学导引 / 期末大作业

A course final project on source-only cosine-to-WSD loss-curve prediction, treating MPL as a frozen baseline and separating transferable LR-drop response from MPL-LD nuisance drift.一份关于 source-only cosine-to-WSD 损失曲线预测的课程期末项目，把 MPL 作为冻结基线，并把可迁移的 LR-drop 响应从 MPL-LD 漂移中分离出来。

MLfinal

Final Project / Selected Topics in Deep Learning Theory深度学习理论选讲 / 期末大作业

A course final project and reproducible PyTorch harness for grokking on modular arithmetic, polynomial modular operations, S5 group tasks, and K-ary modular summation.一份关于 grokking 的课程期末项目和可复现实验框架，覆盖模运算、多项式模任务、S5 群任务和 K-ary 模求和。

Modelmid

Midterm Project / Mathematical Modeling数学模型 / 期中作业

A mathematical modeling midterm project on identifying the source of math solutions, using LaTeX structure, proof rhythm, formula density, and adversarial rewriting as modeling signals.一份关于数学解答来源识别的数学建模期中项目，把 LaTeX 结构、证明节奏、公式密度和对抗改写作为建模信号。

Quant Transformer

Challenge Work / Lingjun Quant Challenge灵均量化挑战赛 / 项目实践

A Lingjun Quant Challenge project using causal Transformer modeling on intraday microstructure data, with emphasis on chronological splits, normalization, and leakage-aware validation.一个灵均量化挑战赛项目，用 causal Transformer 建模日内微观结构数据，重点放在时间切分、归一化和防止信息泄漏的验证流程上。

MultiagentFinal

Final Project / Multi-Agent Foundations多智能体系统 / 期末大作业

A course final project on StrictCoop-Sokoban, intent-grounded communication, and recurrent MAPPO for partially observable cooperative multi-agent reinforcement learning.一份关于 StrictCoop-Sokoban、意图接地通信和 recurrent MAPPO 的多智能体系统期末大作业，研究部分可观测协作任务中的长期角色承诺。

Tuvalon

TA Project Design / Data Structures and Algorithms B数据结构与算法 B / 助教大作业设计

A Flask-based Avalon agent battle platform from pkulab409/pkudsa.avalon, with Python submissions, referee-controlled phases, automatic matching, an ELO system, and replay logs.一个来自 pkulab409/pkudsa.avalon 的 Flask 阿瓦隆智能体对战平台，支持 Python 提交、裁判控制阶段、自动匹配、ELO 系统和对局回放日志。

Uncle Ducky

Independent Mini-Game / Chat-Based Business Narrative独立小游戏 / 聊天式经营叙事

An independent mini-game that uses a chat interface to turn unread messages, supplier pressure, public sentiment, and platform risk into narrative state.一个独立小游戏，用聊天界面把未读消息、供应商压力、舆论和平台风险转化为叙事状态。