diffIRM combines a causal-feature mask, a diffusion-based environment augmentor, and an invariance penalty to improve out-of-distribution spatiotemporal graph prediction on SafeGraph, PeMS04, and PeMS08.
As the relation between饾憣 and causal features饾憢cau is invariant across environments, including causal features can better lower the loss function
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
diffIRM: A Diffusion-Augmented Invariant Risk Minimization Framework for Spatiotemporal Prediction over Graphs
diffIRM combines a causal-feature mask, a diffusion-based environment augmentor, and an invariance penalty to improve out-of-distribution spatiotemporal graph prediction on SafeGraph, PeMS04, and PeMS08.