Yudong Will Xu

I am currently pursuing my PhD in Mechanical and Industrial Engineering at the University of Toronto, under the guidance of Professor Scott Sanner and Professor Elias Khalil. My academic journey began at the same institution, where I completed my Bachelor of Applied Science (BASc) in Engineering Science majoring in Machine Intelligence followed by a Master of Applied Science (MASc) from the Department of Mechanical and Industrial Engineering, also supervised by Professor Sanner and Professor Khalil.

Email  /  CV  /  Google Scholar  /  LinkedIn  /  Twitter  /  Github

My current research focuses on developing machine learning models for solving constraint reasoning problems. I am particularly interested in generative modeling approaches, including diffusion methods and Transformer-based architectures. My work explores how these models can align their reasoning capabilities with the structure of constraints to generate valid and high-quality solutions. My Master's research focused on solving the Abstraction and Reasoning Corpus (ARC-AGI) using object-centric approaches.

Combining blocked Gibbs sampling and diffusion for solving constraint optimization problems.

Composing discrete diffusion models at the factor level to improve compositional reasoning.

Adapting iterative neural constraint heuristics as Large Neighbourhood Search.

Using graph hypernetworks to directly generate full neural-network weights.

Iteratively solving constraint reasoning problems with self-supervised Transformers.

Enhancing the vision transformer (ViT) architecture to solve the Abstraction and Reasoning Corpus.

An exploration of LLMs' ability to solve the Abstraction and Reasoning Corpus.

A graph-based symbolic AI approach to solving the Abstraction and Reasoning Corpus.

Outside of research, I am an avid classical music listener and concertgoer. You can read about some of my experiences and thoughts in the Toronto Star and the Toronto Symphony Orchestra. I have also served as a Volunteer Captain at the Toronto International Film Festival (TIFF) since 2025.

This template is from Jon Barron's website. Here is the source code.