Module III: Neural Knowledge Graphs

Transitioning from "Hard Logic" to "Soft/Statistical Reasoning," this module covers geometric deep learning on graphs.

Knowledge Graph Embeddings (KGE):

Translational Models: TransE (h+r≈t) and its limitations.

Rotational Models: RotatE, which uses rotations in a complex plane to model symmetry and inversion.

Graph Neural Networks (GNNs):

Neural Message Passing: The iterative process of aggregating neighbor information to update node features.

Relational Architectures: R-GCNs and Graph Attention Networks (GATs).

Theoretical Limits: Understanding the Weisfeiler-Lehman (WL) Limit and why standard GNNs may fail to distinguish certain graph topologies.