Gengchen Mai (gengchen.mai@austin.utexas.edu), Assistant Professor, Department of Geography and the Environment, University of Texas at Austin; Ni Lao (nlao@google.com), Senior Research Scientist, Google LLC; Xiaobai Angela Yao (xyao@uga.edu), Professor, Department of Geography, University of Georgia; Nemin Wu (Nemin.Wu@uga.edu), PhD Student, Department of Geography, University of Georgia; Qian Cao (qian.cao1@uga.edu), PhD Student, Department of Geography, University of Georgia; Zhangyu Wang (zhangyuwang@ucsb.edu), Ph.D. Student, Department of Geography, University of California, Santa Barbara
We will be hosting a tutorial on TorchSpatial, a cutting-edge learning framework and benchmark for spatial representation learning (SRL). SRL is essential for learning general-purpose neural network representations from various types of spatial data (e.g., points, polylines, polygons, networks, images, etc.) in their native formats. Learning good spatial representations is fundamental for various downstream applications such as species distribution modeling, weather forecasting, trajectory generation, population estimation, sustainability prediction, geographic question answering, etc.
Although SRL has become the foundation of almost all geospatial artificial intelligence (GeoAI) research, the field has lacked significant efforts to develop a comprehensive deep learning framework and benchmark to support SRL model development and evaluation. To fill this gap, we will introduce TorchSpatial, a learning framework and benchmark for location (point) encoding, one of the most fundamental data types of spatial representation learning. TorchSpatial contains three key components:
1) A unified location encoding framework that consolidates 15 commonly recognized location encoders, ensuring scalability and reproducibility of the implementations;
2) LocBench, a set of benchmark tasks encompassing 7 geo-aware image classification and 4 geo-aware image regression datasets;
3) A comprehensive suite of evaluation metrics to quantify geo-aware models' overall performance as well as their geographic bias, with a novel Geo-Bias Score metric.
This tutorial session will provide geospatial and data science communities with knowledge and tools on how TorchSpatial can help in developing location encoding models and utilizing them for various tasks.
Session Agenda:
1. Introduction to Spatial Representation Learning and Location Encoding
Presenter: Gengchen Mai
Duration: 10 min
2. Overview of TorchSpatial: A Comprehensive Framework for SRL
Presenter: Nemin Wu and Qian Cao
Duration: 8 min
3. TorchSpatial Evaluation: Geo-Bias
Presenter: Zhangyu Wang
Duration: 8 min
4. Hands-on Case Studies with TorchSpatial
Presenter: Nemin Wu and Qian Cao
Duration: 20 min
5. Q&A Session
Duration: 10 minutes
6. Conclusion and Closing Remarks
Presenter: Gengchen Mai
Duration: 4 minutes
Organizers