Lunjun Zhang

I am a PhD student in the Machine Learning Group at University of Toronto, advised by Prof. Raquel Urtasun.

I also work as a researcher at Waabi to develop self-driving technology for long-haul trucking.

I did undergrad in Engineering Science from University of Toronto (2021), during which I spent my summers interning at Vector Institute, Montréal Institute of Learning Algorithms, and Uber Advanced Technologies Group.

Email  /  Google Scholar  /  Twitter  /  Github

profile photo

I work on unsupervised learning for robotics.

I am fascinated by the following question: how can robots autonomously learn from unlabeled experience in unstructured environments, with no external supervision at all?

More specifically, I study what objectives intelligent agents should optimize to learn from the widest range of experience, and how those objectives should be optimized such that neural scaling laws can emerge, meaning that the model performance automatically improves given more (unlabeled) data and compute.

In the context of robotics, I think a lot about unsupervised perception, control, and planning.

Towards Unsupervised Object Detection from LiDAR Point Clouds
Lunjun Zhang, Anqi Joyce Yang, Yuwen Xiong, Sergio Casas, Bin Yang, Mengye Ren, Raquel Urtasun
Conference on Computer Vision and Pattern Recognition (CVPR), 2023

"Can we design unsupervised learning algorithms that discover objects from raw streams of sensor data on their own?"
Understanding Hindsight Goal Relabeling from a Divergence Minimization Perspective
Lunjun Zhang, Bradly Stadie
NeurIPS 2022 Deep RL workshop / Foundation Models for Decision Making workshop

"Can we mathematically describe hindsight relabeling and goal reaching as an imitation learning process from a divergence minimization perspective?"
World Model as a Graph: Learning Latent Landmarks for Planning
Lunjun Zhang, Ge Yang, Bradly Stadie
International Conference on Machine Learning (ICML), 2021 (Long Talk)
paper / poster / code / website

"How can we learn world models that endow agents with the ability to do temporally extended reasoning?"