Integrated Spatio-Temporal Diffusion Bottleneck
Background: Graph-based diffusion models for time-series forecasting often address spatial and temporal dependencies separately or through heuristic coupling.
Question / Future Work: There is a critical need for unified architectures that explicitly integrate spatial graph dependencies with temporal dynamics within a single generative framework, as opposed to relying on decoupled modeling. Investigating this integration is essential for capturing complex, non-stationary dynamic graph signals accurately.
Why It Matters: The current decoupling of space and time is a fundamental limitation in generative graph modeling, hindering the capacity to capture high-order dependencies in dynamic signals.
Metadata & Links
- created_at
- 2026-04-24T05:51:55Z