UrbanFM: Scaling Urban Spatio-Temporal Foundation Models
#UrbanFM #Spatio-Temporal Foundation Models #Urban Computing #WorldST #Zero-shot Generalization #Machine Learning #Urban Data #Scalability
📌 Key Takeaways
- Researchers developed UrbanFM, a scalable urban spatio-temporal foundation model
- The model addresses fragmentation in urban computing through three scaling approaches
- WorldST standardizes data from over 100 global cities into a unified format
- UrbanFM demonstrates zero-shot generalization across unseen cities and tasks
📖 Full Retelling
Researchers led by Wei Chen published a groundbreaking paper titled 'UrbanFM: Scaling Urban Spatio-Temporal Foundation Models' on arXiv on February 24, 2026, introducing a novel approach to urban computing that addresses the fragmentation caused by scenario-specific models that overfit to particular regions or tasks. The research tackles a fundamental challenge in urban systems, which generate continuous spatio-temporal data streams encoding human mobility and city evolution patterns. While foundation models have revolutionized fields like genomics and meteorology, urban computing has remained fragmented due to models designed for specific scenarios rather than general urban principles. The researchers approach this problem systematically by investigating what to scale and how to scale urban spatio-temporal foundation models, identifying three critical dimensions: heterogeneity, correlation, and dynamics. Their solution includes WorldST, a billion-scale corpus standardizing diverse physical signals from over 100 global cities; MiniST units that discretize continuous spatio-temporal fields into learnable computational units; and UrbanFM, a minimalist self-attention architecture designed to learn dynamic dependencies from massive data. The researchers also established EvalST, the largest-scale urban spatio-temporal benchmark to date, demonstrating that UrbanFM achieves remarkable zero-shot generalization across unseen cities and tasks, marking a significant advancement in urban computing.
🏷️ Themes
Artificial Intelligence, Urban Computing, Foundation Models, Spatio-Temporal Data
📚 Related People & Topics
Machine learning
Study of algorithms that improve automatically through experience
Machine learning (ML) is a field of study in artificial intelligence concerned with the development and study of statistical algorithms that can learn from data and generalize to unseen data, and thus perform tasks without explicit instructions. Within a subdiscipline in machine learning, advances i...
Urban computing
Urban computing is an interdisciplinary field which pertains to the study and application of computing technology in urban areas. This involves the application of wireless networks, sensors, computational power, and data to improve the quality of densely populated areas. Urban computing is the techn...
Entity Intersection Graph
Connections for Machine learning:
🌐
Artificial intelligence
5 shared
🌐
Large language model
4 shared
🌐
Reinforcement learning
4 shared
🏢
OpenAI
3 shared
🌐
Review article
1 shared
Original Source
--> Computer Science > Machine Learning arXiv:2602.20677 [Submitted on 24 Feb 2026] Title: UrbanFM: Scaling Urban Spatio-Temporal Foundation Models Authors: Wei Chen , Yuqian Wu , Junle Chen , Xiaofang Zhou , Yuxuan Liang View a PDF of the paper titled UrbanFM: Scaling Urban Spatio-Temporal Foundation Models, by Wei Chen and 4 other authors View PDF Abstract: Urban systems, as dynamic complex systems, continuously generate spatio-temporal data streams that encode the fundamental laws of human mobility and city evolution. While AI for Science has witnessed the transformative power of foundation models in disciplines like genomics and meteorology, urban computing remains fragmented due to "scenario-specific" models, which are overfitted to specific regions or tasks, hindering their generalizability. To bridge this gap and advance spatio-temporal foundation models for urban systems, we adopt scaling as the central perspective and systematically investigate two key questions: what to scale and how to scale. Grounded in first-principles analysis, we identify three critical dimensions: heterogeneity, correlation, and dynamics, aligning these principles with the fundamental scientific properties of urban spatio-temporal data. Specifically, to address heterogeneity through data scaling, we construct WorldST. This billion-scale corpus standardizes diverse physical signals, such as traffic flow and speed, from over 100 global cities into a unified data format. To enable computation scaling for modeling correlations, we introduce the MiniST unit, a novel split mechanism that discretizes continuous spatio-temporal fields into learnable computational units to unify representations of grid-based and sensor-based observations. Finally, addressing dynamics via architecture scaling, we propose UrbanFM, a minimalist self-attention architecture designed with limited inductive biases to autonomously learn dynamic spatio-temporal dependencies from massive data. Furthermore, we establish...
Read full article at source