Explore graph neural networks (GNNs) and their variants in this 35-minute lecture, focusing on their representational power and limitations. Gain insights into the flexible and powerful framework GNNs provide for graph analysis, while understanding their constraints. Survey key results in this field, including the Weisfeiler-Lehman test, WL-hierarchy, and message passing GNN limitations. Delve into high-order WL tests, universal approximation results, and other relevant discussions to enhance your understanding of GNNs and their practical applications in graph analysis.
Yusu Wang - Graph Neural Network: Representational Power and Limitations
Applied Algebraic Topology Network via YouTube
Overview
Syllabus
Introduction
Neural networks for different types of data
General MPNN framework
A simpler pictorial view of MPNN
A simple concrete instantiation
Weisfeiler-Lehman (WL) test
WL-hierarchy and message passing GNN
Limitation of message passing NNs
High-order WL test
Universal Approximation Results
Other Discussions
Taught by
Applied Algebraic Topology Network
