ABSTRACT
Spatial representation learning (SRL) refers to a set of techniques that use deep neural networks (DNNs) to encode and featurize various types of spatial data in the forms of points, polylines, polygons, graphs, etc. In this chapter, we discuss the existing works, key challenges, and uniqueness of spatial representation learning on various types of spatial data. We argue that, as a subfield of spatially explicit artificial intelligence, SRL is a unique research topic that distinguishes GeoAI research and highlights the unique challenges of developing AI models for geospatial data.
