2026
RADE: Unbiased Random Add-Drop Edge as a Regularizer
ICML 2026poster
Graph Neural Networks (GNNs) are prone to overfitting and over-squashing of long-range information. Stochastic graph perturbations (e.g., edge/node dropping) regularize training, but often (i) induce train-test mismatch in expected message aggregation, (ii) lack a principled mechanism for random edg…