.png){kind=small}
Departmental Research Seminars Series
Bridging AI and Geography: Transforming Vision-Language Models through Spatial Intelligence
Date: 19 FEB 2025 (Wednesday)
Time: 10:30-12:00 (HKT)
Venue: CLL, Department of Geography, 10/F, The Jockey Club Tower, Centennial Campus, HKU
Registration link: https://hkuems1.hku.hk/hkuems/ec_hdetail.aspx?guest=Y&ueid=98755
Abstract:
As AI advances, the fusion of multimodal data (e.g., images, text, and spatial information) unlocks unprecedented opportunities for geospatial understanding. This talk introduces an innovative GeoAI framework that extracts and integrates spatial knowledge using Vision-Language Models (VLMs), including GPT-4 and Contrastive Language-Image Pretraining (CLIP), to enhance geospatial analytics. By enabling AI to interpret both visual and linguistic data with spatial awareness, this work expands the capabilities of AI to tackle complex geospatial tasks, such as image geo-localization, land use detection, and urban perception prediction.
The core of this framework leverages spatially explicit prompt engineering and contrastive learning, refining AI models to reason about geo-locations and spatial patterns. These techniques set a new benchmark for AI in geospatial applications, offering scalable, accurate, and explainable insights across diverse environments. As we push the boundaries of GeoAI, the integration of VLMs and spatial reasoning is paving the way for more intelligent, interpretable, and impactful decision-making in the geospatial domain and beyond.
Dr. Meiliu WU
Assistant Professor, GIScience at the University of Glasgow (UoG)
Dr. Meiliu Wu is a Lecturer (Assistant Professor) in GIScience at the University of Glasgow (UoG), where she directs the MSc in Geospatial Data Science and AI. She earned her PhD in Geography from the University of Wisconsin-Madison in May 2024 and joined UoG in August 2024.
Dr. Wu’s research interests include geospatial AI (GeoAI), urban analytics, and environmental sustainability, with her work featured in leading journals and conferences, including JAG, CEUS, Cities, Remote Sensing, and ACM SIGSPATIAL. Her research is pioneering in developing GeoAI foundation models and multimodal learning frameworks, integrating diverse geospatial data sources to advance AI with spatial intelligence.
Profile webpage: https://www.gla.ac.uk/schools/ges/staff/meiliuwu/
Comments