TEMPORAL GRAPH INTELLIGENCE: A CONTINUAL LEARNING FRAMEWORK FOR EVOLVING NETWORK STRUCTURES, EVENT-DRIVEN ANOMALY DETECTION, AND DECENTRALIZED KNOWLEDGE FUSION DEVELOPING SELF-ADAPTIVE NETWORK SCIENCE

The dynamic nature of complex data ecosystems presents significant challenges for conventional graph learning methods. This paper introduces Temporal Graph Intelligence (TGI), a novel framework that leverages Spatio-Temporal Graph Transformers (STGT), continual learning paradigms, and multiscale pattern discovery for robust analysis of evolving network structures. TGI addresses three key challenges: (1) evolving network topologies, (2) real-time anomaly detection in event streams, and (3) decentralized learning in distributed environments. Our approach facilitates self-adaptive network science by integrating spatio-temporal context and hierarchical temporal dependencies. We evaluate TGI on synthetic and real-world datasets, demonstrating significant improvements over baseline models in anomaly detection accuracy, temporal reasoning, and resilience to concept drift.