ICLR ML4RS

I presented this work at the ICLR 2026 ML4RS (Machine Learning for Remote Sensing) Workshop, which focuses on bridging the gap between research and real-world applications. The work addresses the challenge of mapping environmental contamination at scale, where ground truth data is sparse and the target signal is not directly observable from satellite imagery. We introduce FOCUS, a geospatial learning framework that leverages proxy signals and models spatially varying label reliability to learn under noisy supervision. This enables the generation of more reliable contamination risk maps, supporting applications such as identifying high-risk regions and guiding targeted environmental monitoring.

Apr 27, 2026 5:00 PM

Spotlight Oral at ICLR ML4RS