Hi, I'm Zhengxu
- I am an AI Researcher at Huawei London Research Centre. Previously, I was at the Apsara Lab (formerly City Brain Lab, DAMO Academy), Alibaba Cloud, Alibaba Group. I received my Ph.D. from Zhejiang University, under the supervision of Prof. Deng Cai and Prof. Xiaofei He.
- What I’ve always wanted to build is an intelligent system that can explore, learn, and improve by itself, rather than one that only does what it’s told. My research focuses on continual learning, reinforcement learning, and self-improving agents. Here is my DBLP page.
Recent Blogs
深度理解“A Fix for Softmax and Cross-Entropy”
Paper Notes
多智能体 RL 的信用分配
Paper Notes
Latest News
Selected Publications
The Value of Thinking: Learning When to Reason in Hybrid Decision Models
arXiv preprint
Z. Yu, K. Xu, M. Fang, H. Ammar, W. Luo, J. Wang
Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
arXiv preprint
H. Zhao, Z. Yu, Z. He, M. Fang, R. Tutunov, H. Bou-Ammar, W. Luo, J. Wang
From Skills to Talent: Organising Heterogeneous Agents as a Real-World Company
arXiv preprint
Z. Yu, Y. Fu, Z. He, Y. Huang, K. Y. Lee, M. Fang, W. Luo, J. Wang
ChainPrune: Evaluating and Reducing Redundancy in Long Chain-of-Thought Reasoning
arXiv preprint
W. Pan, Z. Yu, Y. Zhang, Z. Jin, B. Lin, X. He, J. Ye
TokenSqueeze: Performance-Preserving Compression for Reasoning LLMs
NeurIPS'25
W. Pan, Z. Yu, Y. Liu, Z. Jin, B. Lin, X. He, J. Ye
Progressive Transfer Learning
TIP'22
Z. Yu, Z. Jin, L. Wei, J. Huang, D. Cai, X. He, X. Hua
MaCAR: Urban Traffic Light Control via Active Multi-agent Communication
IJCAI'20
Z. Yu, S. Liang, L. Wei, Z. Jin, J. Huang, D. Cai, X. He, X. Hua