Laboratory for Deep Structured Learning

The University of British Columbia, Department of Electrical and Computer Engineering

Welcome to the Renjie Liao’s lab for Deep Structure Learning (DSL) at UBC!

Our long-term research goals revolve around:

  1. Perception: building machines that can learn structures (abstraction) from data
  2. Reasoning: building machines that can perform automated inference
  3. Understanding deep learning

In particular, our current research areas involve:

  • Deep Generative Models
  • Geometric Deep Learning
  • Math/Algorithmic Reasoning with Transformers (e.g., LLMs)
  • Probabilistic Inference
  • Visual Understanding and Reasoning
  • Motion Prediction and Planning in Self-driving
  • AI for Healthcare

news

Jul 9, 2024 Renjie was invited to give a Lecture Graph Neural Networks at the CIFAR DLRL Summer School 2024.
Jun 17, 2024 Our work SwinGNN: Rethinking Permutation Invariance in Diffusion Models for Graph Generation was accepted in Transactions on Machine Learning Research (TMLR) 2024. Congratulations to Qi!
May 1, 2024 Our work Learning Latent Structures in Network Games via Data-Dependent Gated-Prior Graph Variational Autoencoders was accepted in International Conference on Machine Learning (ICML) 2024. Congratulations to Xue!
Apr 12, 2024 Renjie was invited to give a talk Graph Neural Networks meet Spectral Graph Theory: A Case Study in the AI seminar at Simon Fraser University.
Apr 9, 2024 Our work An Information-Theoretic Framework for Out-of-Distribution Generalization was accepted in IEEE International Symposium on Information Theory (ISIT) 2024. Congratulations to Wenliang!
Mar 8, 2024 Renjie was invited to give a Distinguished Talk (virtual) Graph Neural Networks meet Spectral Graph Theory: A Case Study at the Vector Institute for Artificial Intelligence.
Feb 29, 2024 Congratulations to Zike for receiving UBC Four Year Doctoral Fellowship (4YF) 2024!

latest posts

selected publications

2024

  1. ICML
    Learning Latent Structures in Network Games via Data-Dependent Gated-Prior Graph Variational Autoencoders
    Xue Yu, Muchen Li, Yan Leng, and Renjie Liao
    In International Conference on Machine Learning (ICML), 2024
  2. ISIT
    An Information-Theoretic Framework for Out-of-Distribution Generalization
    Wenliang Liu, Guanding Yu, Lele Wang, and Renjie Liao
    In IEEE International Symposium on Information Theory (ISIT), 2024
  3. TMLR
    Swingnn: Rethinking permutation invariance in diffusion models for graph generation
    Qi Yan, Zhengyang Liang, Yang Song, Renjie Liao, and Lele Wang
    Transactions on Machine Learning Research (TMLR), 2024
  4. ICLR
    Memorization Capacity of Multi-Head Attention in Transformers
    Sadegh Mahdavi, Renjie Liao, and Christos Thrampoulidis
    In International Conference on Learning Representations (ICLR) Spotlights (5%), 2024
  5. CVPR
    Generative 3D Part Assembly via Part-Whole-Hierarchy Message Passing
    Bi’an Du, Xiang Gao, Wei Hu, and Renjie Liao
    In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

2023

  1. ICLR
    Specformer: Spectral Graph Neural Networks Meet Transformers
    Deyu Bo, Chuan Shi, Lele Wang, and Renjie Liao
    In The International Conference on Learning Representations, 2023
  2. TMLR
    GraphPNAS: Learning Probabilistic Graph Generators for Neural Architecture Search
    Muchen Li, Jeffrey Yunfan Liu, Leonid Sigal, and Renjie Liao
    Transactions on Machine Learning Research (TMLR), 2023
  3. TMLR
    Towards Better Out-of-Distribution Generalization of Neural Algorithmic Reasoning Tasks
    Sadegh Mahdavi, Kevin Swersky, Thomas Kipf, Milad Hashemi, Christos Thrampoulidis, and Renjie Liao
    Transactions on Machine Learning Research, 2023

2022

  1. arXiv
    Gaussian-Bernoulli RBMs without Tears
    Renjie Liao, Simon Kornblith, Mengye Ren, David J Fleet, and Geoffrey Hinton
    arXiv preprint arXiv:2210.10318, 2022
  2. arXiv
    Learning Latent Part-Whole Hierarchies for Point Clouds
    Xiang Gao, Wei Hu, and Renjie Liao
    arXiv preprint arXiv:2211.07082, 2022