Speaker Bio

Caleb is a Principal Research Science Manager in the Microsoft AI for Good Research Lab. He graduated from the Georgia Institute of Technology with a PhD in 2020 and his work focuses on tackling large scale problems at the intersection of remote sensing and machine learning/computer vision. At the AI for Good Lab he co-leads the Geospatial ML research group and is the lead researcher on the Global Renewables Watch, rapid damage assessment, and global building density estimation teams. Caleb is interested in research topics that facilitate using remotely sensed imagery more effectively in conservation, sustainability, and damage response application. For example: self-supervised methods for training deep learning models with large amounts of unlabeled satellite imagery, human-in-the-loop methods for creating and validating modeled layers, and domain adaptation methods for developing models that can generalize over space and time.

Title: Applied GeoML: From Local to Global

Abstract: This talk presents how the Microsoft AI for Good Lab builds geospatial machine learning models and workflows that range from scene level rapid disaster response to global monitoring products. The first part focuses on hyper-local modeling for building damage assessment, where the goal is not broad generalization but fast, scene-specific adaptation to generate actionable maps. The second part shifts to global modeling, where the core challenge becomes validation and robustness across geographies, and time. I’ll highlight three global efforts we work on — Fields of The World, Global Renewables Watch, and TEMPO building density mapping — and discuss what changes when you scale: how you define deployment metrics, how you validate reliably at scale, and how you detect and manage out-of-distribution behavior.