X-AddGraph Provides Post-Hoc Explainability for Dynamic Graph Anomaly Detection
August 14, 2026
X-AddGraph introduces Dual Spatial-Temporal Attribution to explain edge-level anomalies in the AddGraph GCN+GRU framework. It uses gradient-based relevance for adjacency structures and attention weights for temporal context to provide auditable reasoning for flagged edges.
HOW THIS AFFECTS YOU
●
builderYou can add transparency to graph-based anomaly detection systems in regulated environments.
●
researcherThis provides a method for architecture-aligned attribution in recurrent graph neural networks.