acceptodds
Desk rejected at ICLR 2027

OT on the Map: Quantifying Domain Shifts in Geographic Space

Desk rejected at ICLR 2027, so left out of the paper lists and the map.

14 views

Abstract

In computer vision and machine learning for geographic data, out-of-domain generalization is a pervasive challenge, arising from uneven global data coverage and distribution shifts across geographic regions. Though models are frequently trained in one region and deployed in another, there is no principled method for determining when this cross-region adaptation will be successful. Location embeddings, globally available representations of geographic characteristics such as terrain and environmental conditions, can provide a basis for comparing geographic domains. Using location embeddings, we propose a novel strategy for predicting geospatial domain transfer performance with Optimal Transport methods (GeoSpOT). In our experiments, GeoSpOT distances emerge as effective predictors of cross-domain transfer difficulty. We further demonstrate that location embeddings provide information comparable to image/text embeddings, despite relying solely on longitude-latitude pairs as input. This allows users to predict how well a geospatial model trained in a given geographic domain will perform out-of-distribution, even when the exact downstream task is unknown, or no task-specific data is available. Building on these findings, we show that GeoSpOT can also guide the selection of source data for transfer to a target geographic region, without requiring any target-domain input features or labels.

Primary area:
applications to physical sciences (physics, chemistry, biology, etc.)
Keywords:
geospatial machine learning, geographic distribution shift, domain adaptation, transfer learning, optimal transport

Related papers

Discussion (0)

Sign in to comment.