Temporal Sheaf Neural Networks with Dynamic Orthogonal Transport
Researchers have introduced Temporal Sheaf Neural Networks (TSNN), a novel framework for temporal link prediction. Unlike existing models that use a global embedding space, TSNN employs dynamic local frames for each node to capture evolving interaction semantics. This approach ensures causality and preserves hidden states during frame updates, leading to improved performance on various link prediction benchmarks, particularly those with heterogeneous node roles. AI
IMPACT Introduces a new temporal graph modeling technique that improves link prediction accuracy, especially in heterogeneous networks.