Wenhao Li (李 文灏)
I’m a PhD student in Mechanical and Industrial Engineering at the University of Toronto, advised by Prof. Scott Sanner and Prof. Elias Khalil. I hold a Bachelor of Computer Science (Honours) from the University of Waterloo.
I believe programs are taking on two new forms: natural-language instructions executed by LLMs, and neural networks whose architectures and weights encode executable behavior. I study how these programs can be learned from data through world modelling—and how they can recombine existing knowledge to generalize beyond training. Our work on Structure-Aware Graph Hypernetworks demonstrates one path forward.
Before returning to academia, I co-founded AIH Technology and Veriplum, leading AI R&D and technology development. That builder’s perspective keeps my research ambitious, grounded, and executable.
news
| Jul 15, 2026 | Our new paper, OPINE-World, solves 20 of 25 ARC-AGI-3 games with a 78.4 action-efficiency score. Read the project blog and join the conversation on X or LinkedIn. |
|---|---|
| May 13, 2026 | Received an ICML 2026 Gold Reviewer recognition for reviewing service. |
| Apr 22, 2026 | Our paper Structure-Aware Graph Hypernetworks for Neural Program Synthesis was published at ICLR 2026. View the poster. |
| Apr 10, 2026 | Gave an invited talk on our paper Certifiable Boolean Reasoning Is Universal with Hrad Ghoukasian (McMaster University) as part of the Fields Institute’s Mathematical AI Seminar. Talk details. |
selected publications
- Structure-Aware Graph Hypernetworks for Neural Program SynthesisIn The Fourteenth International Conference on Learning Representations, 2026
-
- Tackling the Abstraction and Reasoning Corpus with Vision Transformers: the Importance of 2D Representation, Positions, and ObjectsTransactions on Machine Learning Research, 2025