About Me
My name is Weida Wang (王蔚达), and I also go by David. I am currently a first-year Ph.D. student at the College of Computer Science and Artificial Intelligence, 
Fudan University, advised by Prof. Wanli Ouyang. I also conduct AI for Science research with the 
Shanghai AI Laboratory.
Prior to this, I received my B.Eng. in Software Engineering from the School of Computer Science and Technology, 
Tongji University in 2025. I worked closely with Prof. Jin Zeng, Prof. Lin Zhang, and Prof. Guanjie Zheng.
My research interests focus on AI Scientist, Agent for Science, RSI, and LLM Reasoning. I am particularly interested in building autonomous systems that can reason, learn, and continually improve through real scientific workflows.
🐕 Collaboration: I am always happy to discuss LLM reasoning and AI for Science. If our interests overlap, please feel free to contact me.
🔥 News
🎉 Our chemical reasoning model Chem-R and molecule generation benchmark S²-Bench were accepted to KDD 2026. See you in Jeju 🇰🇷!
🚀 Our efficient reasoning method Step-GRPO was accepted to ACL 2026. Many thanks to Dr. Mingbao Lin and Prof. Min Zhang.
🎉 Our condensed matter physics benchmark CMPhysBench was accepted to ICLR 2026. Many thanks to all my collaborators. See you in Rio 🇧🇷!
🎉 Our work GIGA-ToF was accepted to ICCV 2025. Many thanks to my collaborators.
📣 Joined the OpenScienceLab at the Shanghai AI Laboratory as a research intern in the AI for Science group.
🎉 Selected for Tongji University’s Outstanding Science and Innovation Talent — Youth Enlightenment Program.
📝 Selected Publication (Full List)
* Equal contribution. † Corresponding author.
Consistent Time-of-Flight Depth Denoising via Graph-Informed Geometric Attention
- GIGA-ToF combines graph-informed geometric attention with motion-invariant fusion to improve both temporal stability and spatial sharpness in depth video.
Step-GRPO: Internalizing Dynamic Early Exit for Efficient Reasoning
- Step-aware reinforcement learning internalizes early exit into the model itself, reducing redundant reasoning without adding an inference-time controller.
TRACK: Temporal Decoupled Kriging for Inductive Spatio-Temporal Graph
- TRACK separates node-specific inherency from dynamic correlations to improve inductive spatio-temporal kriging across traffic and air-quality datasets.
🏅 Honors and Awards
- 2023 National Scholarship (top 0.2% nation-wide)
- 2025 Outstanding Graduate Award of Shanghai
- 2022,2023,2024 Merit Student in Tongji University
- 2023 🥇 Gold Medal of International Genetically Engineered Machine Competition (iGEM) AI & Software Track
- 2023 🥇 First Prize of National Undergraduate Mathematics Competition (Non-mathematics Category)
- 2023 🥇 First Prize of China Undergraduate Computer Design Competition (top 0.1% nation-wide)
- 2023 🥈 Second Prize of China Collegiate Computing Contest HCI Innovation Competition (top 0.4% nation-wide)
- 2024 🥉Third Prize of China Collegiate Computing Contest Mobile Application Innovation Contest (top 1% nation-wide)
- 2023 🥇 First Prize of HuaShu Cup National Undergraduate Mathematical Modeling Contest (top 2% nation-wide)
- 2023 🥈 Honorable Mention of Mathematical Contest in Modeling
- 2022 🥇 First Prize of Undergraduate Mathematics Competition (Non-mathematics Category) in Shanghai
📖 Educations
- 2025-2030, College of Computer Science and Artificial Intelligence, Fudan University, Shanghai, China
- Supervised by Prof. Wanli Ouyang
- 2021-2025, School of Computer Science and Technology, Tongji University, Shanghai, China
- Supervised by AP. Jin Zeng and Prof. Lin Zhang
💻 Internships
- 2024.10 - Present, AI for Science Group, Shanghai AI Laboratory [link]
- 2023.07 - 2025.03, Graph Signal Processing Lab, Tongji University
- 2023.11 - 2024.08, Intelligent Internet of Things Research Center (IIOT), Shanghai Jiao Tong University [link]
- 2023.11 - 2024.03, City Science Lab@Shanghai (MIT Media Lab) [link]
- 2023.03 - 2024.03, Key Laboratory of Geotechnical and Underground Engineering of the Ministry of Education, Tongji University [link]
- 2023.07 - 2023.08, Hundsun Technologies Inc. [link]
🧩 Projects
ImagiTale
An AI-assisted interactive storybook for children, built for iPad with SwiftUI, speech interaction, visual segmentation, and personalized feedback.
CCCC HCI 2023 · Best Innovation Award · National Second Prize
CASleuth — the Virus Detective
An educational game and bioinformatics tool for CRISPR-Cas virus detection, created with the Tongji-Software iGEM team.
iGEM 2023 · Gold Medal · Software & AI Track