Explore the fundamentals and advanced concepts of graph mining in this comprehensive 43-minute conference talk from KDD2020. Dive into the world of learning on graph-structured data, covering essential topics such as node classification, link prediction, and graph classification in real-world scenarios. Discover the rise of deep learning on graphs and understand the goal of graph representation learning. Examine key challenges in graph data and the limitations of traditional node embeddings. Investigate the transition from node embeddings to graph neural networks (GNNs) and their impact on drug repurposing and recommender systems. Delve into the connections between GNNs, graph convolutions, and Graph Fourier analysis. Learn about the Weisfeiler-Lehman algorithm and explore practical GNNs beyond the WL hierarchy. Address the challenges of message passing and gain insights into breaking its bottleneck. This talk by William L. Hamilton provides a comprehensive overview of graph mining techniques and their applications in various domains.
Overview
Syllabus
Intro
Learning on graph-structured data
Node classification in the real world
Link prediction in the real world
Graph classification in the real world
The rise of deep learning on graphs
Outline for today
What is graph representation learning Goal: Learning useful node and graph representations without hand-crafted features.
Key challenges of graph data . Two key challenges with graph-structured data
Limitations of traditional node embeddings
Node embeddings to graph neural networks
What about graph classification?
The impact of GNNs: drug repurposing
The impact of GNNs: recommender systems
GNNs and graph convolutions
Graph Fourier analysis
What does this have to do with GNNS
Weisfeiler-Lehman algorithm
Practical GNNs beyond the WL hierarchy
Breaking the bottleneck of message passing
Three challenges or one?
Taught by
Association for Computing Machinery (ACM)
