Spatiotemporal Imputation with Graph-Informed Flow Matching

In three linesGiFlow, a graph-informed flow matching framework, addresses missing data imputation in spatiotemporal systems (air quality, urban traffic). It replaces standard Gaussian priors with graph-informed priors via spatiotemporal filtering, and uses a hybrid vector field model combining spatial attention, temporal attention, and spatiotemporal propagation. Code available.

Summary generated by Claude — human-verified