Training and Deploying Pre-trained Remote Sensing Models

Abstract: In this talk, I will share my experience pre-training and deploying remote sensing models. Ultimately, I hope these models can be useful to people solving important problems. I'll discuss how this goal, and users' needs (such as accessibility and spatial generalization), have influenced the design of models I helped develop (Presto, Galileo, OlmoEarth). I'll share lessons from putting these models into production, including deploying Presto into the WorldCereal system.  Bio: Gabi is a research scientist at Ai2 on the OlmoEarth team. He completed a PhD at Mila under the supervision of David Rolnick, focussing on machine learning for remote sensing. This included working closely with Hannah Kerner and NASA Harvest, NASA's agriculture and food security program.