GeoMAE: Masking Representation Learning for Spatio-Temporal Graph Forecasting with Missing Values
Researchers have developed GeoMAE, a novel self-supervised learning model designed to handle missing data in spatio-temporal graph forecasting. This model addresses limitations in existing methods by incorporating dynamic spatial correlations and improving generalizability across varied missing data patterns. GeoMAE utilizes an attention-based network and a masking autoencoder approach, demonstrating significant performance improvements over current benchmarks on real-world datasets. AI
IMPACT Introduces a new method for handling missing data in spatio-temporal forecasting, potentially improving urban intelligence systems.