About Me
Mathematics · Artificial Intelligence · Systems for thought
I am a second-year undergraduate student in Mathematics and Artificial Intelligence at the Beijing Institute of Mathematical Sciences and Applications (BIMSA). I am learning how mathematical ideas, machine intelligence, and carefully designed workflows can help people reason better under uncertainty.
My current direction is exploratory but intentional: I am building small systems, reading across learning theory and language models, and looking for research problems where rigorous modeling meets real human decision-making. I am especially grateful to Prof. Rongling Wu, who introduced me to learning theory and encouraged me to look for structure behind empirical results.
About
I am transitioning from being mainly a student of existing ideas to becoming a researcher-builder: someone who studies foundations, implements systems, and learns from the friction between theory and use.
My interests sit between three habits of mind:
- Mathematical discipline — using probability, optimization, and learning theory to make vague intuitions testable.
- System thinking — treating models, data, users, incentives, and interfaces as one coupled system rather than isolated components.
- Personal experimentation — building tools for attention management, cognition, and human-computer workflows, then reflecting on how they change behavior.
At this stage, I care less about presenting a finished identity and more about developing a durable research taste: asking precise questions, building minimal prototypes, and keeping a long memory of what works and what fails.
Research Interests
Learning theory and sequential decision-making
Multi-armed bandits, online learning, reinforcement learning theory, generalization, and how agents learn from partial feedback over time.
Language models for economic reasoning
Using natural language models together with quantitative models to parse narratives, extract signals, and study decision-making in financial and economic settings.
Quantitative systems
Designing research pipelines that connect data collection, feature construction, backtesting, uncertainty estimation, and model evaluation.
Cognition and human-computer workflows
Tools that help people manage attention, externalize thought, remember context, and collaborate with intelligent systems without losing agency.
Projects and Building
I am currently building projects as a way to discover sharper questions. I prefer small, inspectable systems that make assumptions visible and can grow into research prototypes.
Language-driven quantitative research pipeline
Problem. Economic information often appears first as text: reports, news, commentary, and changing market narratives.
Approach. Explore how language models can structure unstructured text, connect it with quantitative signals, and support careful hypothesis testing instead of ad-hoc prediction.
Next. Build evaluation routines that separate useful signal from overfitting, hindsight bias, and narrative convenience.
Attention and cognition workflow experiments
Problem. Modern work produces too much context switching; good ideas are often lost between notes, tasks, and unfinished reading.
Approach. Prototype lightweight workflows for capturing questions, preserving context, and turning reading or coding sessions into reusable knowledge.
Next. Test which routines actually improve focus and long-term recall, rather than only feeling productive.
Learning theory reading map
Problem. It is easy to collect papers without building a coherent mental model of a field.
Approach. Maintain a structured map of concepts across bandits, online learning, reinforcement learning, and deep learning theory.
Next. Turn notes into short technical essays and reproducible toy examples.
Recent News
- Nov 2025 Began participating in the BIMSA Mathematics and Artificial Intelligence Program.
- Now Exploring research directions at the intersection of NLP, quantitative modeling, learning theory, and human-centered AI workflows.
Notes and Philosophy
Build small systems, keep the assumptions visible, and let repeated contact with reality refine the question.
I want this homepage to remain a living research log rather than a static résumé. As my work matures, I plan to add short notes on papers, project write-ups, technical experiments, and reflections on building better tools for thought.
Professional Services
- Journal Reviewing: Currently none
- Conference Reviewing: Currently none