陈宇卓Yuzhuo Chen
Yuzhuo Chen

你好，我是Hi, I'm

陈宇卓Yuzhuo Chen Yuzhuo Chen陈宇卓

让 AI 理解、生成，并与世界交互Teaching AI to understand, generate, and interact with the world

兴趣：多模态生成 · 多模态理解 · 机器人 · 仿真 · 游戏 · 强化学习Interests: multimodal generation · multimodal understanding · robotics · simulation · games · RL

📍 中国科学技术大学（USTC） · 中国·合肥 📍 University of Science and Technology of China (USTC) · Hefei, China

✉ 联系我✉ Get in touch 谷歌学术Google ScholarGitHubGitHub领英LinkedInORCIDORCID
查看全部论文 →All publications →

我好奇的问题与探索Questions I'm Exploring

查看全部探索 →All explorations →

记录持续感兴趣的技术路线与研究疑问：认真阅读已有工作，也区分个人判断和经过验证的结论。Technical interests and open questions, separating the literature, hypotheses, and personal perspectives.

INTEREST · 技术路线INTEREST · Technical survey

如何让模型学会自适应推理How Can Models Learn Adaptive Reasoning?

为什么自适应推理影响 Agent 的固定预算有效产出与机器人在线时延？梳理 SFT、成本敏感 RL、步骤级信用分配、反应—规划切换的最新研究；区分真正学会思考时机与仅减少 CoT，并讨论有限推理机会能否促进参数化直觉的待验证假设。A critical survey of SFT, cost-aware RL, reasoning credit assignment, adaptive planning and robot latency. Tests the distinction between shorter CoT and genuinely learned, transferable parametric intuition.

阅读技术路线与个人思考 →Read the survey and perspective →

INTEREST · 技术路线INTEREST · Technical survey

一个 Agent 是如何组织推理、工具调用与跨轮状态的？How Do Agents Organize Reasoning, Tool Calls, and Cross-Turn State?

从模型 Token、API 结构化输出到 Harness 的工具执行和状态回传，比较 Claude Messages、OpenAI Responses/Codex、Qwen/vLLM、OpenHands、LangGraph、smolagents 与 OpenCode；解释 Thinking、Tool Call、执行反馈和跨轮历史分别由谁负责。An end-to-end comparison of how Claude, OpenAI Responses/Codex, Qwen/vLLM, OpenHands, LangGraph, smolagents and OpenCode organize reasoning, tool use, execution feedback and conversation state.

阅读技术路线与个人思考 →Read the survey and perspective →

OPEN QUESTION · 机器人学习OPEN QUESTION · Robot Learning

目前的 Tokenizer，是否天然具备数值连续性？Do Today's Tokenizers Inherently Preserve Numerical Continuity?

普通语言 Tokenizer 不保证相近数值有相近表示；VLM 的连续视觉特征也不等于坐标输出具备连续性，而离散 Action Codebook 同样可能扭曲物理邻近关系。对照 xVal、BitTokens、Pix2Seq、FAST、X-Tokenizer、ActionPiece、ProAct 和连续动作头，提出几种可验证的改进路径。Text tokenization does not inherently preserve numerical distance. Continuous visual features do not guarantee metric output continuity, and discrete VLA action codes may distort physical neighborhoods. Review numerical encodings, action codecs, continuous heads, and testable research directions.

查看文献和可验证实验 →Explore evidence and experiments →

继续探索Explore more

📚

研究知识库Research Library

文献综述 · 术语 · 学习路线 · 研究问题Surveys, concepts, reading paths, questions
📝

论文成果Publications

公开论文与学术成果Published papers and scholarly work
🛠️

项目与工具Projects & Tools

研究实现、开源工具与演示Research implementations and open-source tools
✍️

写作与笔记Writing

技术解释、研究随笔与教程Explanations, research notes, tutorials

了解我的研究与实习经历Explore my research and internship experience

查看经历 →View experience →

👋 个人简介👋 Profile

我是中国科学技术大学网络空间安全学院的二年级硕士研究生， 导师为张卫明教授与俞能海教授。 此前在山东大学控制科学与工程学院获得学士学位。我的研究聚焦于人工智能、计算机视觉与安全的交叉领域。

欢迎交流研究思路、开源工具和潜在合作。如果有共同感兴趣的问题，欢迎通过邮件联系。

I am a second-year M.S. student at the School of Cyber Science and Technology, University of Science and Technology of China (USTC), advised by Prof. Weiming Zhang and Prof. Nenghai Yu. Previously I received my B.S. from Shandong University. My research lies at the intersection of AI, computer vision, and security.

I'm happy to discuss research questions, open-source tools, and potential collaborations. Feel free to reach out by email.

生成式 AI（AIGC）Generative AI (AIGC)计算机视觉Computer Vision信息隐藏Information HidingAI 安全AI SafetyAI + 医疗AI for Medicine

🔥 近况🔥 News

  • 2026.10.11 📊 发布 ICLR 2027 投稿分析：43,013 篇投稿翻倍之后，同行评审的把关机制还跟得上吗？📊 Released an analysis of 43,013 ICLR 2027 submissions: can peer review keep up with a doubling of submissions?
  • 2026.10.01 🛠️ 开源 Cursor Remote Lite：在手机上接着用电脑里的 Cursor。🛠️ Open-sourced Cursor Remote Lite: use the Cursor on your computer from your phone.
  • 2025.11.08 🎉 一篇论文被 AAAI 2026 接收。🎉 One paper accepted by AAAI 2026.
  • 2025.06.26 🎉 一篇论文被 ICCV 2025 接收。🎉 One paper accepted by ICCV 2025.

进入研究知识库 →Explore the research library →