Learning Graphs from Data

Speaker:  Renjie Liao, The University of British Columbia.

Time:       11:00 am, Aug. 1^st

Location: SIST1A 200

Host:        Xuming He

Abstract:

This talk will mainly present two approaches pushing the boundaries of graph generation and network games. First, SwinGNN, a non-invariant diffusion model, addresses the challenges of permutation-equivariant networks by leveraging edge-to-edge 2-WL message passing and shifted window-based self-attention. Key training and sampling techniques enhance graph generation quality, with a post-processing method ensuring permutation invariance. Extensive experiments on synthetic and real-world datasets like molecule generation demonstrate SwinGNN's state-of-the-art performance. Second, the Data-Dependent Gated-Prior Graph Variational Autoencoder (GPGVAE) uncovers latent structures in network games by inferring interaction types and network structures from observed actions. GPGVAE employs a spectral Graph Neural Network encoder and a data-dependent gated prior, complemented by a Transformer-based mixture of Bernoulli encoders and a GNN-based decoder. Systematic experiments highlight GPGVAE's effectiveness in inferring network structures and capturing interaction types. I will also mention other recent work from our lab.

Bio:

Renjie Liao is an Assistant Professor (since Jan. 2022) in ECE and CS (associated member) Departments at UBC. He is also a Faculty Member at Vector Institute for AI and a Canada CIFAR AI Chair. He was a Visiting Faculty Researcher at Google Brain, working with Geoffrey Hinton and David Fleet. He received his PhD from UofT in 2021, advised by Richard Zemel and Raquel Urtasun. During his PhD, he also worked as a Senior Research Scientist at Uber ATG. He obtained his MPhil from CUHK, advised by Jiaya Jia, and his BEng from Beihang. His research focuses on geometric deep learning, deep generative models, and their intersections with computer vision and self-driving.