ECHO: Elastic Speculative Decoding with Sparse Gating for High-Concurrency Scenarios
International Conference on Machine Learning (ICML) Spotlight, 2026
Yuhao Shen (沈宇豪)
Hey, I'm Yuhao Shen, a direct PhD student at the College of Control Science and Engineering, Zhejiang University, advised by Prof. Cong Wang. I also received my B.E. degree from Zhejiang University.
My research lies in MLSys, LLM Inference, AI Infrastructure, and Edge Computing. Over the past two years, I have been deeply engaged in the field of speculative sampling and decoding. I was previously a research intern at Qwen Application and received an internship offer from the Tencent Hunyuan Qingyun Project. Currently, I am researching RL rollout acceleration in the Tongyi Qwen Foundation Model Infra group. Outside academia, I enjoy playing basketball, video games, and photography.
Recent News
- Congratulations, ECHO is chosen as ICML 2026 Spotlight!
- Congratulations, Double is chosen as ACL 2026 Best Paper Candidate!
- Double and Spark are accepted to ACL 2026 Main Conference, and Atlas is accepted to ACL 2026 Findings!
- Two papers, Vision-TTT and ECHO, are submitted to arXiv!
- ParallelVLM is accepted at CVPR 2026!
- SpecBranch is accepted at ICLR 2026!
- Multiple papers including Double, Atlas, SSL, Spark, and TALON submitted to arXiv!
- Hetero2 Pipe is accepted at ICDCS 2025!
- I started my Direct Ph.D. journey at Zhejiang University!
Education
|
Zhejiang University, Hangzhou, China Direct Ph.D. in Control Science and Engineering, 2024 - 2029 Advisor: Cong Wang |
|
|
Zhejiang University, Hangzhou, China Bachelor of Engineering in Control Science and Engineering, 2020 - 2024 Advisor: Cong Wang |
Publications
Atlas: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning
ACL 2026 Findings
Hetero2 Pipe: Pipelining Multi-DNN Inference on Heterogeneous Mobile Processors under Co-Execution Slowdown
IEEE 45th International Conference on Distributed Computing Systems (ICDCS), 2025
SENGraph: A self-learning evolutionary and node-aware graph network for soft sensing in industrial processes
IEEE Transactions on Neural Networks and Learning Systems, 2024
Vision-TTT: Efficient and Expressive Visual Representation Learning with Test-Time Training
arXiv (2026)
Profession
|
Research Intern, Qwen Foundation Model May. 2026 - Present Topic: Speculative Decoding, RL Infra Advisor: Yucheng Li, Huiqiang Jiang Hangzhou, China |
|
|
Research Intern, Qwen Application Nov. 2025 - May. 2026 Topic: Speculative Decoding, AI Infra Advisor: Ye Shuang, Jun Dai, Lei Chen Hangzhou, China |