Knowledge Graph Reasoning with Graph Neural Networks

Explore knowledge graph reasoning with graph neural networks in this 26-minute talk by Zhaocheng Zhu from the University of Montreal and Mila. Dive into the fundamental problem of predicting answers to queries by reasoning over existing facts in knowledge graphs. Learn about two innovative approaches: Neural Bellman-Ford Networks (NBFNet) for single-hop reasoning and Graph Neural Network Query Executor (GNN-QE) for multi-hop queries. Discover how these models relate to traditional symbolic methods while addressing missing links in knowledge graphs. Gain insights into the visualization of intermediate reasoning steps, enhancing understanding of the process. Cover topics such as inductive settings, path representations, efficient computation, first-order logic queries, and fuzzy logic operations. Understand the applications of knowledge graphs in natural language understanding, recommender systems, and drug discovery.

Knowledge Graphs
Knowledge Graph Reasoning
Inductive Setting
Path Representations
Efficient Computation
Generalized Bellman-Ford Algorithm
Examples
Neural Parameterization
Interpretation on FB15k-237
First-Order Logic Queries
Symbolic Methods
Neural Methods
Four Operations
Relation Projection
Fuzzy Logic Operations