AI for Science: Improving Prediction Through Data Uncertainty and Rare Change Detection

How can we quantify uncertainty to improve AI-based prediction and identify rare change signals? This talk explores these questions through image and time-series analysis. I will discuss how prediction margins and entropy guide active learning to reduce labeling effort, and how class ambiguity indices quantify data-level uncertainty to help explain model predictions and errors. Finally, I will introduce AI methods for learning pixel trajectories to detect rare changes in time-series images. Environmental applications will illustrate how these methods advance AI for science through more efficient learning and interpretable prediction.

Bio
https://search.asu.edu/profile/1978357

Dr. Wenwen Li is a Foundation Professor and Director of the Spatial Analysis Research Center (SPARC) in the School of Geographical Sciences and Urban Planning at Arizona State University (ASU). She also leads the Cyberinfrastructure and Computational Intelligence (CICI) Lab. Dr. Li's research interests include cyberinfrastructure, big data, geospatial artificial intelligence (GeoAI), and their applications in data- and computation-intensive environmental and social sciences. Dr. Li is a Fellow of the American Association of Geographers (AAG) and the University Consortium for Geographic Information Science (UCGIS), and a recipient of the NSF CAREER Award and the NSF Mid-Career Advancement Award.