Yiming Wang(王一鸣)

Ph.D. Candidate @ SJTU, School of Computer Science

No.800, Dongchuan Road, Minhang District, Shanghai, China.

alsaceym@gmail.com

Yiming Wang
Curriculum Vitae

Profile

I was born in Jiangsu Province, China, in August 2001. I am now a third-year Ph.D. candidate at the School of Computer Science, Shanghai Jiao Tong University, supervised by Prof. Rui Wang. I also work closely with Prof. Zhuosheng Zhang. Before that, I received a Bachelor's degree from the Institute of Artificial Intelligence, Beihang University.

Research Interests

My research interests center on two directions:

  • LLM Reasoning: Developing stronger general reasoning capabilities, while making reasoning more interpretable / efficient / multi-dimensional (including multilingual, multimodal, and agentic settings).
  • Self-Evolving LLMs: Developing self-iteration and generalization capability in zero-external-data or zero-supervised-label settings, through approaches including in-context learning, test-time scaling, and reinforcement learning.

Education

2023.09 - Present Department of Computer Science and Engineering, Shanghai Jiao Tong University (SJTU), Shanghai, China
2019.09 - 2023.06 Institute of Artificial Intelligence, Beihang University (BUAA), Beijing, China

Internships

2024.07 - 2025.07 Qwen Team, Alibaba, Hangzhou
Research Intern: Multilingual / Multimodal LLMs
Mentor: Dr. Baosong Yang
2023.03 – 2023.09 Institute of AI Industry Research (AIR), Tsinghua University, Beijing
Research Intern: AI for Science
Mentor: Prof. Hao Zhou

Technical Report

Qwen3 Technical Report

One of Contributors

  • Introduces the Qwen3 model family, unifying thinking and non-thinking modes while advancing reasoning, efficiency, and multilingual capabilities across dense and MoE architectures.

Selected Publications [Full Publications]

PolyMath: Evaluating Mathematical Reasoning in Multilingual Contexts

Yiming Wang, Pei Zhang, Jialong Tang, Haoran Wei, Baosong Yang, Rui Wang, Chenshu Sun, Feitong Sun, Jiran Zhang, Junxuan Wu, Qiqian Cang, Yichang Zhang, Fei Huang, Junyang Lin, Fei Huang, Jingren Zhou

Annual Conference on Neural Information Processing Systems (NeurIPS), 2025

  • Introduces PolyMath, a multilingual mathematical reasoning benchmark spanning 18 languages and four difficulty levels for evaluating reasoning models across languages and complexities.
  • Incorporated into the official benchmark of the Qwen3 family.

Sampling-Efficient Test-Time Scaling: Self-Estimating the Best-of-N Sampling in Early Decoding

Yiming Wang, Pei Zhang, Siyuan Huang, Baosong Yang, Zhuosheng Zhang, Fei Huang, Rui Wang

Annual Conference on Neural Information Processing Systems (NeurIPS), 2025 Spotlight

  • Proposes Self-Truncation Best-of-N (ST-BoN), which identifies promising reasoning paths from early hidden-state consistency and truncates inferior samples without a reward model.

Latent Space Chain-of-Embedding Enables Output-free LLM Self-Evaluation

Yiming Wang, Pei Zhang, Baosong Yang, Derek F. Wong, Rui Wang

International Conference on Learning Representations (ICLR), 2025

  • Proposes Chain-of-Embedding, a training-free and output-free method that uses progressive hidden-state trajectories to estimate the correctness of LLM responses.

RetroDiff: Retrosynthesis as Multi-stage Distribution Interpolation

Yiming Wang, Yuxuan Song, Yiqun Wang, Minkai Xu, Rui Wang, Hao Zhou, Wei-Ying Ma

International Conference on Artificial Intelligence and Statistics (AISTATS), 2025

  • Formulates retrosynthesis as multi-stage distribution interpolation and develops a conditional diffusion model that sequentially generates external groups and bonds from products.

Embedding Trajectory for Out-of-Distribution Detection in Mathematical Reasoning

Yiming Wang, Pei Zhang, Baosong Yang, Derek F. Wong, Zhuosheng Zhang, Rui Wang

Annual Conference on Neural Information Processing Systems (NeurIPS), 2024

  • Proposes the Trajectory Volatility score, which detects out-of-distribution mathematical reasoning inputs from layer-wise embedding-shift trajectories.

Meta-Reasoning: Semantics-Symbol Deconstruction For Large Language Models

Yiming Wang, Zhuosheng Zhang, Pei Zhang, Baosong Yang, Rui Wang

Findings of the Association for Computational Linguistics: ACL 2024

  • Introduces Meta-Reasoning, which abstracts semantically different reasoning tasks into shared symbolic meta-forms to improve in-context reasoning and generalization.

Element-aware Summarization with Large Language Models: Expert-aligned Evaluation and Chain-of-Thought Method

Yiming Wang, Zhuosheng Zhang, Rui Wang

Annual Meeting of the Association for Computational Linguistics (ACL), 2023

  • Builds expert-written element-aware evaluation sets and proposes Summary Chain-of-Thought to generate more comprehensive and human-aligned summaries.

Academic Service

Reviewer

  • ICML (2026 – ), NeurIPS / ICLR / AAAI (2025 – ), ACL ARR Rolling / IEEE TASLP (2024 – )

Teaching Assistant (SJTU)

  • Natural Language Processing and Large Language Model (2024 – 2025, for the John Class)

Teaching Assistant (BUAA)

  • Cognitive Basis (2022)
  • Advanced Linear Algebra (2020 – 2022, Outstanding TA for Algebra Courses)