M2Net for Spatial-Temporal Forecasting via Scale-Specific Masking
October 20, 2026
M2Net unifies scale-specific structural learning and semantic decomposition to improve spatial-temporal forecasting. It utilizes a Multi-Scale Masked attention mechanism to generate scale-adaptive spatial masks, addressing granularity-dependence and signal entanglement in urban data.
HOW THIS AFFECTS YOU
●
researcherThis method provides a way to decouple global patterns from node-specific disturbances in time-series data.