Author(s): Yoshihiro Shibuo

Linked Author(s):

Keywords: Graph-based optimization; Soft shortest-path; Water infrastructure networks

Abstract: This study proposes a machine-learning framework that learns latent edge costs and obtains minimal-cost pathways through an entropy-regularized soft shortest-path solver. The method integrates a learnable edge-cost model with differentiable flow optimization, enabling structure extraction without prescribed labels. Numerical experiments on a lattice grid show that the learned costs lead to concentrated soft-flow patterns and quantized routes consistent with minimal-cost behavior. The approach provides a basis for extending cost-driven network-structure inference to a broad range of water-infrastructure network problems.

DOI:

Year: 2026