Gengchen Mai
University of Texas at Austin, TX·Computer Science
AI models that work with maps, satellite images and location data often perform well in some parts of the world and poorly in others. This unevenness is called geographic bias, and it matters because people in under-served regions may get less reliable results from tools built on these models.
- Fields
Computer Science
Earth & Environmental Science
Mathematics & Statistics
- Topics
geographic bias
GeoAI
geospatial artificial intelligence
- Grants
- 1 known grant, through Aug 2028
What they work on
- Develop geographic bias metrics grounded in spatial point pattern analysis, statistics describing how points are distributed across space, that handle continuous space and give comparable scores across GeoAI and foundation models.
- Build a geographic bias evaluation benchmark that combines diverse geospatial datasets, tasks and the new metrics, to serve as a quality and ethics check for geospatial AI and foundation models.
- Create lightweight debiasing algorithms that reduce a foundation model's geographic bias at the inference stage, when the model is used, without retraining it.
- Develop pre-training-stage debiasing methods for foundation models that reduce geographic bias and improve how well the models generalize across geospatial tasks.
More about the research
The project asks how to measure this bias fairly and how to reduce it. It covers both task-specific GeoAI (geospatial artificial intelligence) models built for one job and foundation models, large general-purpose AI models adapted to many tasks.
Why it matters
AI tools for mapping, navigation, environmental monitoring and other location-based tasks can quietly serve some regions worse than others. Standard, comparable bias metrics and an open benchmark could let developers in academia and industry check their models for geographic unfairness before deployment. Debiasing methods could help make these tools more reliable for communities in regions that current models handle poorly.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Jundong Li
University of Virginia, VA·Computer Science·Early-career
Design graph neural networks tailored to peptides, including amino acid-aware message passing and graph transformers, and combine them with pretrained protein language models to merge structural and evolutionary information.
Jurij Leskovec
Stanford University, CA·Computer Science
Develop multi-level graph neural networks that handle graphs whose nodes are themselves complex structures, such as molecular graphs, for scientific foundation models.
Kasim Candan
Arizona State University Campus Immersion, AZ·Computer Science
Develop PanAX methods that separate setting-specific from transferable parts of epidemic data and models, so knowledge from data-rich outbreaks can be reused for new outbreaks with limited data.
The school
University of Texas at Austin: 49 labs on PI Pages, $1.1B a year on research, 871 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.