Neural Graph Databases: Combining GNNs with Labeled Property Graphs
Scalable Parallel Computing Lab, SPCL @ ETH Zurich via YouTube
Overview
Explore the integration of graph neural networks (GNNs) with graph databases (GDBs) in this 15-minute conference talk from the First Learning on Graphs Conference (LoG'22). Discover how the Labeled Property Graph (LPG) data model can be leveraged to combine the computational capabilities of GDBs with the predictive power of GNNs. Learn about LPG2vec, an encoder that transforms LPG datasets into representations compatible with various GNN models. Understand how this approach preserves rich information from LPG labels and properties, potentially increasing prediction accuracy by up to 34%. Gain insights into the concept of neural graph databases and their potential to enhance graph machine learning methods for complex data analysis.
Syllabus
Introduction: Graph Databases and Labeled Property Graph Data Model
Introduction: GNNs
Marrying Graph Databases and GNNs: Label and Property Prediction
LPG2vec: Encoding LPG Datasets
Evaluation
Taught by
Scalable Parallel Computing Lab, SPCL @ ETH Zurich