STA-GANN: A Valid and Generalizable Spatio-Temporal Kriging Approach

TL;DR

STA-GANN is a novel GNN-based framework that enhances the validity and generalizability of spatio-temporal kriging, effectively capturing dynamic dependencies and shifts in incomplete data scenarios.

Contribution

It introduces a decoupled phase module, dynamic metadata graph modeling, and adversarial transfer learning to improve spatio-temporal pattern inference and generalization.

Findings

Outperforms existing methods on nine diverse datasets.

Effectively captures dynamic spatial dependencies and temporal shifts.

Demonstrates theoretical and empirical superiority in validity and generalization.

Abstract

Spatio-temporal tasks often encounter incomplete data arising from missing or inaccessible sensors, making spatio-temporal kriging crucial for inferring the completely missing temporal information. However, current models struggle with ensuring the validity and generalizability of inferred spatio-temporal patterns, especially in capturing dynamic spatial dependencies and temporal shifts, and optimizing the generalizability of unknown sensors. To overcome these limitations, we propose Spatio-Temporal Aware Graph Adversarial Neural Network (STA-GANN), a novel GNN-based kriging framework that improves spatio-temporal pattern validity and generalization. STA-GANN integrates (i) Decoupled Phase Module that senses and adjusts for timestamp shifts. (ii) Dynamic Data-Driven Metadata Graph Modeling to update spatial relationships using temporal data and metadata; (iii) An adversarial transfer learning strategy to ensure generalizability. Extensive validation across nine datasets from four fields and theoretical evidence both demonstrate the superior performance of STA-GANN.