Tag
This paper proposes a scalable local Sinkhorn divergence framework for training stochastic neural networks to reconstruct multidimensional random fields, with theoretical generalization error bounds and numerical demonstrations for uncertainty quantification.