Pinzheng Wang | Homepage

📝 Biography

I know it because I built it.

- I am a second-year Ph.D. student at the Institute of Artificial Intelligence, Soochow University, under the supervision of Assoc. Prof. Juntao Li and Prof. Min Zhang.

- Before that, I obtained the Bachelor's degree in Artificial Intelligence from Soochow University (2019–2023), ranking 1st out of 67 and awarded the First-Class Scholarship.

- Currently, my research lie at Reinforcement learning and Large Reasoning Model, including but not limited to Self-Play RL, Self-Correction, Latent Reasoning, Efficient Reasoning.

Check my publications and CV (PDF) for more information!

Seeking research internship opportunities in Reinforcement Learning and LLM Reasoning!

📚 Research Interests

Large-scale Pretraining of Foundation Model

Diffusion-based LLM: Bridge_Gap_Diffusion (ACL2023)
Language Foundation Model: OpenBA V1 & V2 (SCIS 2024)

Knowledge Editing

Over-attention in knowledge editing: SADR (ICLR 2025)

Reinforcement learning

Self-play RL: CDG (ICML 2025)

📖 Education & Internship

Research Intern, HuaWei Clound BU, 2025.06~current
Research of RL ans reasoning model

Phd Student, Computer Science, 2023-current
Institute of Artificial Intelligence, Soochow University

B.Eng., Computer Science and Technology, 2019-2023
School of Computer Science, Soochow University

🔥 News

May, 2025. We release a self-play RL method, CDG — Critic Discernment Game Training, to improve rationality in the reasoning process of language models. Our work has been accepted at the ICML 2025 main conference.

Paper:

Project:

Feb, 2025. We reveal the over-attention issue in knowledge eiditing and proprose a general method, SADR — Critic Discernment Game Training, to siginificantly mitigate specificity failure. Our work has been accepted at the ICLR 2025 main conference.

Paper:

Project:

📒 Selected Publications

Check my Google Scholar for more information! (* stands for equal contribution)

[ICML 2025 (CCF A)] Improving Rationality in the Reasoning Process of Language Models through Self-playing Game
Pinzheng Wang, Juntao Li, Zecheng Tang, Haijia Gui, Min Zhang
[ICLR 2025 (CAAI A)] Revealing and Mitigating Over-attention in Knowledge Editing
Pinzheng Wang, Zecheng Tang, Keyan Zhou, Juntao Li, Qiaoming Zhu, Min Zhang
[Preprint 2024] OpenBA-V2: Reaching 77.3% High Compression Ratio with Fast Multi-Stage Pruning
Dan Qiao*, Yi Su*, Pinzheng Wang*, Jing Ye, Wenjing Xie, Yuechi Zhou, Yuyang Ding, Zecheng Tang, Jikai Wang, Yixin Ji, Yue Wang, Pei Guo, Zechen Sun, Zikang Zhang, Juntao Li, Pingfu Chao, Wenliang Chen, Guohong Fu, Guodong Zhou, Qiaoming Zhu, Min Zhang
[SCIS 2024 (CCF A)] OpenBA: An Open-sourced 15B Bilingual Asymmetric seq2seq Model Pre-trained from Scratch
Juntao Li*, Zecheng-Tang*, Yuyang Ding*, Pinzheng Wang*, Pei Guo, Wangjie You, Dan Qiao, Wenliang Chen, Guohong Fu, Qiaoming Zhu, Guodong Zhou, Min Zhang
[ACL 2023 Long (CCF A)] Bridging the Gap between Training and Inference for Diffusion Model
Zecheng Tang*, Pinzheng Wang*, Keyan Zhou, Juntao Li, Ziqiang Cao, Min Zhang

🎖️ Scholarship & Honors

First-Class Scholarship, ZhongYuanKeDa Scholarship, Soochow University, Jun 2022.
Rank 1/67, School of Computer Science and Technology
CCF CCSP East China Regional Gold Award (Top 10%)

CCPC Jiangsu Provincial Contest, Silver Medal. .