ADMFormer: An Adaptive-Decomposition Transformer with Time-Varying Masked Spatial Attention for Traffic Forecasting
Researchers have developed ADMFormer, a novel Transformer-based model designed for more accurate traffic forecasting. This model addresses challenges in traffic data by first decomposing signals into stable periodic patterns and event-driven fluctuations. It then uses a dual-branch temporal module to process these components separately and a time-varying masked spatial attention mechanism to dynamically focus on relevant spatial dependencies. Experiments show ADMFormer outperforms existing methods on real-world traffic datasets. AI
IMPACT Introduces a novel architecture for improved traffic forecasting, potentially enhancing intelligent transportation systems.