SCOT: Multi-Source Cross-City Transfer with Optimal-Transport Soft-Correspondence Objective

TL;DR

SCOT introduces a novel optimal transport-based framework for multi-source cross-city transfer learning, explicitly modeling soft region correspondences to improve prediction accuracy in heterogeneous, label-scarce urban environments.

Contribution

It proposes a Sinkhorn-based optimal transport method for explicit soft region matching and a balanced transport approach for multi-source transfer, enhancing robustness and interpretability.

Findings

SCOT consistently improves transfer accuracy across real-world city datasets.

The learned transport couplings offer interpretable diagnostics of city-region alignments.

SCOT enhances robustness in heterogeneous urban transfer tasks.

Abstract

Cross-city transfer improves prediction in label-scarce cities by leveraging labeled data from other cities, but it becomes challenging when cities adopt incompatible partitions and no ground-truth region correspondences exist. Existing approaches either rely on heuristic region matching, which is often sensitive to anchor choices, or perform distribution-level alignment that leaves correspondences implicit and can be unstable under strong heterogeneity. We propose SCOT, a cross-city representation learning framework that learns explicit soft correspondences between unequal region sets via Sinkhorn-based entropic optimal transport. SCOT further sharpens transferable structure with an OT-weighted contrastive objective and stabilizes optimization through a cycle-style reconstruction regularizer. For multi-source transfer, SCOT aligns each source and the target to a shared prototype hub using balanced entropic transport guided by a target-induced prototype prior. Across real-world cities and tasks, SCOT consistently improves transfer accuracy and robustness, while the learned transport couplings and hub assignments provide interpretable diagnostics of alignment quality.