Hidden Axis of Uncertainty: Latent-Posterior Alignment in Graph Neural Networks with Bayesian Output Layers
This work observes that predictive uncertainty decreases as latent representations shift toward lower-variance posterior directions, even though the posterior variance does not contract, and proposes Alignment-Guided Learning (AGL), which explicitly promotes this alignment during training.
S. Choi, Damdae Park, Junhyuk Choi et al.
· 0 citations