CV

Yuyang Hu

📧 yuyang.hu@ruc.edu.cn
📞 (+86) 133-0917-7613
🔗 GitHub: https://github.com/namespace-ERI
🎓 Google Scholar: https://scholar.google.com/citations?user=_t-3ipgAAAAJ

Education

Renmin University of China (RUC), Beijing
PhD Student, Gaoling School of Artificial Intelligence
2025 – Present
Expected graduation: June 2030

Renmin University of China (RUC), Beijing
Bachelor of Science, Gaoling School of Artificial Intelligence
2021 – 2025
GPA: 3.63 / 4.0

Research Interests

  • Large Language Models (LLMs)
  • Long-Horizon Agents
  • Self-Evolving Agents, Agent Memory, and AutoResearch Agents
  • Information Retrieval

Experience

Shanghai Artificial Intelligence Laboratory, Shanghai
Research Intern, LLM Center
August 2026 – Present
Mentor: Prof. Tao Gui
Research focus: ML Coding

Publications

  • Repo-To-Skill: Distilling GitHub Repositories Into AI4AI Skills
    arXiv, 2026
    Jianlyu Chen*, Yuyang Hu*, Hongjin Qian*, Jiawei Liu*, Wenqing Wei*, Xiaolong Chen et al.
    arXiv

  • AREX: Towards a Recursively Self-Improving Agent for Deep Research
    arXiv, 2026
    Shuqi Lu, Chaofan Li, Kun Luo, Zhang Zhang, Hui Wang et al., Yuyang Hu, et al.
    arXiv

  • OPOD: On-Policy Omni Distillation
    arXiv, 2026
    Tong Zhao, Yuyang Hu, Reed Li, Yu Lu, Haibo Shi, Yutao Zhu, Zhicheng Dou
    arXiv

  • Towards Long-Horizon Agents: A Survey
    Preprints, 2026
    Guanting Dong, Xiaoshuai Song, Yuyang Hu, Jiajie Jin, Chenghao Zhang et al.
    OpenReview

  • VeriGraph: Towards Verifiable Data-Analytic Agents
    arXiv, 2026
    Jiajie Jin, Zhao Yang, Wenle Liao, Yuyang Hu, Guanting Dong, Xiaoxi Li, Yutao Zhu, Zhicheng Dou
    arXiv

  • Toward Generalist Autonomous Research via Hypothesis-Tree Refinement
    arXiv, 2026
    Jiajie Jin*, Yuyang Hu*, Kai Qiu, Qi Dai, Chong Luo et al.
    arXiv

  • From Player to Master: Enhancing Test-Time Learning of LLM Agents via Reinforcement Learning over Memory
    arXiv, 2026
    Yishuo Cai, Xingyu Guo, Xuancheng Huang, Jinhua Du, Can Huang et al., Yuyang Hu, et al.
    arXiv

  • From Prompt Injection to Persistent Control: Defending Agentic Harness Against Trojan Backdoors
    arXiv, 2026
    Jiejun Tan, Zhicheng Dou, Xinyu Yang, Yuyang Hu, Yiruo Cheng, Xiaoxi Li, Ji-Rong Wen
    arXiv

  • AgentFugue: Agent Scaling for Long-Horizon Tasks through Collective Reasoning
    arXiv, 2026
    Yuyang Hu, Hongjin Qian, Shuting Wang, Jiongnan Liu, Tong Zhao, Xiaoxi Li, Zheng Liu, Zhicheng Dou
    arXiv

  • SAM: State-Adaptive Memory for Long-Horizon Reasoning Agent
    arXiv, 2026
    Yuyang Hu, Hongjin Qian, Shuting Wang, Jiongnan Liu, Ziliang Zhao, Jiejun Tan, Zheng Liu, Zhicheng Dou
    arXiv

  • MemSifter: Offloading LLM Memory Retrieval via Outcome-Driven Proxy Reasoning
    arXiv, 2026
    Jiejun Tan, Zhicheng Dou, Liancheng Zhang, Yuyang Hu, Yiruo Cheng, Ji-Rong Wen
    arXiv

  • Memory in the Age of AI Agents
    arXiv, 2025; HuggingFace Daily #1
    Yuyang Hu*, Shichun Liu*, Yanwei Yue*, Guibin Zhang*, Boyang Liu et al.
    arXiv

    A comprehensive survey and unifying framework for memory in AI agents. We systematize existing concepts and paradigms, propose a three-dimensional analysis framework (Forms, Functions, and Dynamics), summarize representative benchmarks and open-source frameworks, and discuss future directions including multi-agent memory and integration with reinforcement learning.

  • Memory Matters More: Event-Centric Memory as a Logic Map for Agent Searching and Reasoning
    ACL 2026 Findings
    Yuyang Hu, Jiongnan Liu, Jiejun Tan, Yutao Zhu, Zhicheng Dou
    arXiv

    We propose CompassMem, an event-centric memory framework that moves beyond memory as a passive external database. Memory is organized into events and connected via explicit logical relations to form an event graph, enabling goal-oriented memory retrieval and long-horizon reasoning.

  • Pretrain Once, Finetune Repeatedly: Toward Reusable Pretrained Models for Generative Retrieval
    Under Review (SIGIR 2026)
    Yuyang Hu, Yujia Zhou, Xiaoxi Li, Tong Zhao, Zhicheng Dou

    We investigate cross-domain pretraining for generative retrieval, analyzing the impact of model architecture, parameter scale, and data scale. We propose PreGR, a reusable pretraining framework with a two-stage synthetic query filtering strategy that reduces reliance on domain-specific synthetic data and repeated training.

  • Investigating Users' Search Behavior and Outcome with ChatGPT in Learning-oriented Search Tasks
    SIGIR-AP 2024
    Sijie Liu, Yuyang Hu, Zihang Tian, Zhe Jin, Shijin Ruan, Jiaxin Mao

Skills

  • Programming & Systems: Proficient in Python; experienced with model development, training, and debugging in Linux environments
  • Deep Learning Frameworks: Extensive experience with PyTorch and HuggingFace Transformers; familiar with large-scale model training
  • Models & Algorithms: Strong understanding of Transformer-based architectures, large language models, and agent systems
  • Research Practice: Actively follow and reproduce cutting-edge research in LLMs and AI agents

Honors & Awards

  • Outstanding Graduate of Beijing — June 2025
  • Beijing Merit Student — October 2024
  • RUC First-Class Academic Excellence Scholarship — October 2024
  • RUC Second-Class Academic Excellence Scholarship — October 2023
  • RUC Second-Class Academic Progress Scholarship — October 2022

Miscellaneous

  • Technical Blog (Xiaohongshu): Paper reviews and research notes with a focus on Long-Horizon Agents, Self-Evolving Agents, and Agent Memory
  • English Proficiency:
    • CET-6: 542
    • CET-4: 631