Spotlight: SEAI Lab at UGA
Information provided by Professor Gengchen Mai, director of the Spatially Explicit Artificial Intelligence (SEAI) Lab at the University of Georgia
Dec. 7, 2022
Please briefly introduce yourself and your lab.
I am an assistant professor in the UGA Geography Department and an affiliated professor of the UGA Computer Science Department and the UGA AI Institute. Before coming to UGA, I was a postdoctoral scholar at the Stanford Artificial Intelligence Lab (SAIL), Stanford Computer Science, working with Prof. Stefano Ermon. I received my Ph.D. in GIScience from UCSB Geography in 2021. My research mainly focuses on spatially explicit artificial intelligence (SEAI), GeoAI, geospatial knowledge graphs, deep learning on remote sensing data, and computational sustainability. Until now, I have published 54 peer-reviewed scientific publications in top international journals and CS/GIS conferences. I am the winner of three best paper awards (AGILE 2019, ACM K-CAP 2019 & 2021), and the recipient of various awards such as the Top 10 WGDC 2022 Global Young Scientist Award, AAG's 2022 William L. Garrison Award, a Jack and Laura Dangermond Graduate Fellowship, a UCSB Schmidt Fellowship, a Microsoft AI for Earth Grant, and so on.
As Michael Goodchild put it, “A model is said to be spatially explicit when it differentiates behaviors and predictions according to spatial location”. In the context of machine learning (ML) and artificial intelligence (AI), we propose the idea of spatially explicit artificial intelligence, which focuses on better ways to design ML/AI model architectures for consuming spatial information by incorporating spatial knowledge and spatial inductive bias.
The Spatially Explicit Artificial Intelligence (SEAI) Lab is named after the SEAI idea. I established this lab in Fall 2022. Currently, we have one CS undergraduate student and a geography Ph.D. student. We are looking for 1–2 Ph.D. students next fall.
Describe the overall 5–10 year vision and/or mission of your lab.
The SEAI Lab has three specific missions:
- developing a set of basic and composable spatially explicit AI models as a sharable tool set;
- creating an interlinked open geospatial knowledge graph for various geospatial tasks;
- designing innovative applications of spatially explicit AI models on different geospatial tasks across different domains such as remote sensing, climatology & meteorology, ecology, environmental science, agriculture, urban data science, economy, and so on.
What distinguishes your lab from others working on GeoAI topics?
- Research questions: the SEAI Lab is interested in fundamental GeoAI questions, technical innovation, and algorithm development for GeoAI research.
- Interdisciplinary collaboration network: the SEAI Lab has strong collaborations with different departments and academic institutions including different departments of UGA (CS, Electronic and Computer Engineering, Public Health, Agriculture, and AI Institute), US universities (Stanford CS & Earth Science, UCSB Geography & CS, UC Davis Transportation, UC Berkeley Geography, ASU Geography, UMD Geography, Univ. Buffalo Geography, Univ. Wisc.–Madison Geography, Univ. Washington Geography, Caltech CS, CMU ML), and international universities/institutions (Univ. Bristol, Universiteit Utrecht, Univ. British Columbia, McGill University, ETH, Technische Universität München, National University of Singapore, Nanyang Technological University, Hong Kong University of Science and Technology, Wuhan University, and so on).
- Industry collaborators: we are currently collaborating with research scientists from multiple high-tech companies including Google X, Google AI, Waymo, Esri, Apple, SayMosaic.ai, Meta, OpenAI, Nvidia, Amazon, and so on.
How does your lab’s work address real-world problems?
Our objective is to develop SEAI models that can be utilized in various GeoAI tasks in different domains such as remote sensing (e.g, building change detection), climatology (e.g., precipitation prediction, drought forecasting), ecology (e.g., species distribution estimation), environmental science (e.g., air pollution forecasting), agriculture (e.g., crop yield prediction), urban data science (e.g., traffic forecasting), economy (e.g., poverty prediction), and so on. We have shown the effectiveness of spatially explicit AI in tasks such as geographic question answering, point-of-interest classification, geo-aware species fine-grained recognition, shape classification, wealth index prediction, and so on. Some of our published work has been adopted by different companies (Google X, Apple Maps, Esri, IOS Press) as a part of their production pipeline.
Name one (or more) person or organization whose work has influenced your thinking.
- Dr. Krzysztof Janowicz, Professor, UC Santa Barbara: Jano introduced me to the field of geospatial knowledge graphs and spatially explicit models.
- Dr. Ni Lao, Senior Research Scientist, Google AI: Ni is my industry advisor and introduced me to the general AI field.
- Dr. Stefano Ermon, Associate Professor, Stanford University: Stefano helped to broaden my horizons in ML and AI.
Do you consider your lab/work to be interdisciplinary? If so, how?
Our lab’s work is highly interdisciplinary. We have an interdisciplinary collaboration network. The research problems we study are also from different domains such as remote sensing, agriculture, public health, atmospheric science, economics, ecology, and so on.
Do you consider your lab diverse? In what ways? Are there ways in which you would like to become more (or less) diverse?
Currently, there are two female students working with me. One is a CS major undergraduate student and the other is a geography Ph.D. student. We plan to increase the diversity of the lab in the future.
What’s one thing you wish your [colleagues, students/postdocs, or anyone else] knew about your lab, but don’t (in general)?
We have very strong collaborations with multiple industry sectors such as Google AI, Google X, OpenAI, and so on. Students have a high chance of finding research internships at Google X or other high-tech companies.
What are some notable examples of student/postdoc research topics in your lab?
- Location representation learning
- Self-supervised learning on satellite images
- Zero-shot learning on large language models on geospatial semantics tasks
What backgrounds, in terms of education and/or experience, are useful to have for students/postdocs in your lab?
A bachelor's/master's degree in GIScience, CS, information science, or a related field before the official enrollment (Jan. 2023 or Aug. 2023) is required. Research experience in ML and AI would be a plus but is not required.
What are you looking for in a student/postdoc, and how can they research out to you?
- A bachelor's/master's degree in GIScience, CS, information science, or a related field before the official enrollment (Jan 2023 or Aug 2023).
- Great passion for GIScience and GeoAI research.
- Familiar with one programming language, e.g., Java or Python.
- Familiar with one spatial analysis and mapping tool, e.g., ArcGIS or QGIS.
- Research experience in ML and AI would be a plus, but is not required.
- Fluent in spoken English.
What do you offer these students/postdocs in return, so to speak?
- Students will publish multiple high-quality peer-reviewed papers at top AI conferences such as NeurIPS, ICLR, KDD, ACM SIGAPTIAL, ISWC and top CS/GIScience journals such as IJGIS, TGIS, JOSIS, TKDE, and so on.
- Students will have a high chance of finding research internships at Google X or other high-tech companies. This will prepare them to find industry jobs after graduation or greatly benefit their GeoAI research if they plan to go to academia.
- Competitive salary.
- Detailed guidance on their research.
- Multiple opportunities to attend top international/national AI/ML/GIScience conferences.
- Many collaboration opportunities with various institutions.