Zhenrui Yue

👋 About Me

Hello, I’m Zhenrui Yue (岳真锐 in Chinese), a research scientist at Google DeepMind. My research focuses on building language models and agents that can learn from interaction to solve long-horizon tasks, adapt over time, and ultimately improve themselves through the experience and feedback they generate.

I recently earned my PhD at University of Illinois Urbana-Champaign, where I was advised by Dr. Wang. During my doctoral studies, I also spent time as a student researcher and research intern at Google DeepMind, Meta MSL / GenAI, and NVIDIA. Prior to that, I earned my B.S. and M.S. degrees from Technische Universität München.

🌋 Research Interests

My research lies at the intersection of reinforcement learning for long-horizon tasks, recursive self-improvement, and continual learning. I aim to build language agents that learn through extended interaction, accumulate knowledge and skills as their environments evolve, and ultimately generate the experience and feedback that drive their own improvement. Realizing this vision requires moving beyond today’s LLM training, which often relies on static datasets, sparse rewards, and one-off training pipelines. These limitations make it difficult to assign credit across long trajectories, acquire new capabilities without forgetting existing ones, and sustain reliable, open-ended improvement. To address these challenges, my work and interests span three connected areas:

• Reinforcement learning for long-horizon tasks:
  • Train LLMs to reason in hybrid discrete and continuous spaces (HRPO)
  • Teach agents to reason with search and other external tools (Search-R1)
  • Develop stable and efficient learning signals for long-horizon decision-making, with an emphasis on credit assignment
• Recursive self-improvement:
  • Close the loop between task generation and learning so that agents can create increasingly challenging curricula without human annotation (Dr. Zero)
  • Advance on-policy distillation and self-distillation to complement sparse environmental rewards with dense learning signals and bootstrap successive policy improvements
  • Study how self-generated tasks, verifiable feedback, and persistent memory can enable reliable, open-ended improvement
• Continual learning:
  • Develop continual post-training methods that acquire new knowledge and behaviors while preserving existing capabilities
  • Improve models’ use of computation and evidence at test time through search, reasoning, and test-time scaling (Inference Scaling)
  • Adapt LLMs to emerging domains and tasks by learning from limited supervision & human preferences (RAFTS, RARG)

💼 Experience

Google DeepMind Research Scientist

Mountain View, CA Dec 2025 - Present

Meta MSL / GenAI Research Intern

Menlo Park, CA May 2025 - Dec 2025

Google DeepMind Student Researcher

Mountain View, CA May 2024 - Dec 2024

NVIDIA Research Intern

Remote, IL May 2023 - Aug 2023

📚 Selected Publications

• Dr. Zero: Self-Evolving Search Agents without Training Data
  Zhenrui Yue, Kartikeya Upasani, Xianjun Yang, Suyu Ge, Shaoliang Nie, Yuning Mao, Zhe Liu, Dong Wang
  COLM 2026 [Paper] [Code]

• Hybrid Latent Reasoning via Reinforcement Learning
  Zhenrui Yue, Bowen Jin, Huimin Zeng, Honglei Zhuang, Zhen Qin, Jinsung Yoon, Lanyu Shang, Jiawei Han, Dong Wang
  NeurIPS 2025 [Paper] [Code]

• Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning
  Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan Arik, Dong Wang, Hamed Zamani, Jiawei Han
  COLM 2025 [Paper] [Code]

• Inference Scaling for Long-Context Retrieval Augmented Generation
  Zhenrui Yue*, Honglei Zhuang*, Aijun Bai, Kai Hui, Rolf Jagerman, Hansi Zeng, Zhen Qin, Dong Wang, Xuanhui Wang, Michael Bendersky
  ICLR 2025 (Oral) [Paper]

• Retrieval Augmented Fact Verification by Synthesizing Contrastive Arguments
  Zhenrui Yue, Huimin Zeng, Lanyu Shang, Yifan Liu, Yang Zhang, Dong Wang
  ACL 2024 [Paper] [Code]

• Linear Recurrent Units for Sequential Recommendation
  Zhenrui Yue*, Yueqi Wang*, Zhankui He, Huimin Zeng, Julian McAuley, Dong Wang
  WSDM 2024 [Paper] [Code]

 Full publication list can be found on my google scholar.

✨ Miscellaneous

 SPC / AC: ARR (ACL, EMNLP, NAACL, etc.).

 Reviewer: AAAI, ARR, COLM, ICCV, ICLR, KDD, NeurIPS, TMLR, etc.

 Teaching: Discrete Math, Intro to Database, Machine Learning, Computer Networks.

 Languages: Chinese, Cantonese, English, German and a bit Spanish.

 More will be added when I have time :)

Powered by Jekyll and Minimal Light theme.