Geospatial data is common across a wide range of disciplines and modeling tasks, e.g. in ecology or urban analytics. Location features are often not readily available and need to be obtained via individual data collection and fusion. This opens the opportunity for a new class of "foundation models": global, general-purpose geographic location encoders, which provide vector embeddings summarizing the characteristics of a location for convenient usage in downstream tasks. I will outline the intuition and technical challenges for building these models, and contextualize them with respect to other geospatial foundation models in the vision, language and geophysical domain. Bio: I am a postdoctoral researcher at Microsoft Research New England and part of the Machine Learning and Statistics group. My research focuses on the representation of geographic phenomena in machine learning methods, particularly in neural networks. My recent work includes the integration of notions of spatial dependency into neural networks and the unsupervised training of location encoders that learn characteristics of a given location and can be deployed in different downstream tasks. My work is motivated by real-world challenges such as climate change and increasing urbanisation, combining technical and methodological research with application and deployment studies. I have a PhD in Computer Science and Urban Science from the University of Warwick and spent time as a visiting student at NYU, as an Enrichment student at the Alan Turing Institute and as a Beyond Fellow at TUM and DLR.
