Hello, I'm你好，我是
AI Researcher · Computer Vision & Multimodal LLMsAI 研究员 · 计算机视觉与多模态大模型
I build vision & language models that hold up in the real world.让视觉与语言模型
M.S. student in Electrical & Computer Engineering at Rowan University, advised by Dr. Ying Tang (IEEE Fellow). Co-author of five papers, including two in IEEE Transactions (one as first author) and one at ACM Multimedia. Previously a computer vision intern at Lenovo.罗文大学电子与计算机工程硕士在读，师从唐瑛教授（IEEE Fellow）。合著论文 5 篇，含 IEEE Transactions 期刊论文 2 篇（其中 1 篇为第一作者）、ACM Multimedia 论文 1 篇；曾任联想计算机视觉实习生。
I'm a master's student in Electrical and Computer Engineering at Rowan University, working with Dr. Ying Tang (IEEE Fellow). My research sits where computer vision meets multimodal large language models, with a focus on models that hold up in messy, real-world conditions.
我是罗文大学电子与计算机工程专业的硕士研究生，师从唐瑛教授（IEEE Fellow）。研究方向是计算机视觉与多模态大模型的交叉领域，尤其关注模型在复杂真实场景中的可靠性。
My main project right now is cockpit visual understanding: teaching models to read HUD instruments, sectional charts, weather radar and traffic displays the way a pilot does. In parallel, I build LLM peer agents that personalize game-based learning for students.
目前的主要课题是驾驶舱视觉理解：让模型像飞行员一样读懂 HUD 仪表、航图、气象雷达与交通显示。同时，我也在开发 LLM 同伴智能体，为学生提供个性化的游戏化学习支持。
Before Rowan, I earned a B.S. in Artificial Intelligence from Chongqing Jiaotong University and spent the better part of a year at Lenovo building production OCR and handwriting-recognition pipelines. I like closing the loop: from paper, to prototype, to something people actually use.
此前，我在重庆交通大学完成人工智能专业本科学习，并在联想实习近一年，参与研发生产级 OCR 与手写识别流程。我喜欢把事情做成闭环：从论文到原型，再到真正有人在用的产品。
Two IEEE Transactions journal papers (one as first author), a paper at ACM Multimedia (CCF-A), and an IEEE FIE paper on AI-driven learning.
2 篇 IEEE Transactions 期刊论文（其中 1 篇为第一作者）、1 篇 ACM Multimedia（CCF-A）论文，以及 1 篇 IEEE FIE 教育 AI 论文。
Built online and offline OCR / handwriting-recognition pipelines at Lenovo. I also deploy and run my own AI services end to end: React, FastAPI, Nginx, Linux.
在联想完成在线与离线两条 OCR / 手写识别生产流程；也能独立完成 AI 服务的全栈开发、部署与运维（React、FastAPI、Nginx、Linux）。
Hands-on with HUD symbology, sectional charts, weather radar, ADS-B and TCAS-style traffic logic, territory few ML researchers cover.
深入接触 HUD 显示符号、航图、气象雷达、ADS-B 与类 TCAS 交通显示逻辑，这是少有机器学习研究者涉足的领域。
I design and evaluate LLM peer agents for game-based STEM learning, from prompt design to the interactions students see.
设计并评估面向 STEM 游戏化学习的 LLM 同伴智能体，从提示词设计一直做到学生端的交互体验。
National first prize in English public speaking (FLTRP·ETIC Cup, 2021). Comfortable presenting and collaborating in both English and Chinese.
“外研社·国才杯”全国英语演讲大赛一等奖（2021）。能用中英文自如地汇报、沟通与协作。
I turn ideas into interactive prototypes, so collaborators can try a result instead of only reading about it.
习惯把研究想法做成可交互的原型，让合作者能直接上手体验，而不只是看论文里的插图。
National first prize, English public speaking (2021)全国英语演讲大赛一等奖（2021）
Rowan University · Dr. Ying Tang's Lab罗文大学 · 唐瑛教授课题组·Glassboro, NJ美国新泽西州
Rowan University罗文大学·Personalized Instruction & Need-aware Gamification个性化教学与需求感知游戏化
Lenovo联想·Beijing, China北京
Robust perception under real-world shift, and AI that helps people learn.
关注真实场景偏移下的稳健感知，以及能帮助人学习的 AI。
Test-Time Few-Shot Object Detection via Dynamic Prototype Fusion
IEEE Transactions on Cybernetics2026
Makes few-shot detectors adapt at inference time by dynamically refining class prototypes under domain and category shift.
在域偏移与类别偏移下动态精炼类别原型，让小样本检测器在推理阶段稳健自适应。
TIDE: Test-Time Few-Shot Object Detection
IEEE Transactions on Systems, Man, and Cybernetics: Systems2024
Introduces test-time few-shot detection: no fine-tuning required, enabling fast or black-box deployment.
提出测试时小样本检测设定：无需微调模型，即可支持快速部署与黑盒场景。
RDLNet: A Novel and Accurate Real-world Document Localization Method
ACM International Conference on Multimedia (MM 2024)2024
A real-world document-localization benchmark plus a lightweight network that stays accurate under difficult mobile capture conditions.
提出真实场景文档定位基准与轻量级定位网络，在复杂的移动端拍摄条件下依然保持高精度。
Enhancing Fatigue Detection through Heterogeneous Multi-Source Data Integration and Cross-Domain Modality Imputation
arXiv preprint 2507.168592025
Frames real-world fatigue detection as a multi-source, modality-misaligned learning problem, transferring knowledge from high-fidelity sensors to deployable settings.
将真实场景下的疲劳检测建模为多源、模态不对齐的学习问题，把高精度传感器中的知识迁移到可实际部署的场景。
Enhancing Game-Based Learning with AI-Driven Peer Agents
IEEE Frontiers in Education Conference (FIE 2025), work in progress2025
Studies AI peer agents as in-game teammates and measures whether SPARC-style game-based learning improves student outcomes.
研究 AI 同伴智能体作为游戏内队友的作用，评估 SPARC 式游戏化学习能否切实提升学习效果。
M.S., Electrical & Computer Engineering电子与计算机工程 · 硕士
Advisor: Dr. Ying Tang (IEEE Fellow). Research in aviation visual understanding and educational AI.
导师：唐瑛教授（IEEE Fellow）。研究方向：航空视觉理解与教育 AI。
B.S., Artificial Intelligence人工智能 · 本科
I'm open to AI research roles and to collaborations in computer vision, multimodal LLMs, aviation AI and educational AI. Email is the best way to reach me.
正在寻找 AI 研究相关岗位，也欢迎计算机视觉、多模态大模型、航空 AI 与教育 AI 方向的合作。邮件联系最为方便。