|
NASA Harvest, in partnership with NASA’s Office of the Chief Science Data Officer, MLCommons, and Arizona State University, is launching an exciting new benchmark challenge—hosted on Dynabench—focused on smart training data selection for geospatial ML.
The challenge asks a bold question: Can smaller, smarter training sets outperform larger, unfocused ones for localized predictions? Using the CropHarvest satellite dataset, participants must select data subsets that maximize crop classification accuracy in four regions—China, Togo, Tigray, and Senegal.
Why it matters? Your solution could shape NASA Harvest’s real-world monitoring systems, boosting global food security and disaster readiness.
🗓 Runs April 14 – August 3, 2025
🔍 Focus: Data-centric strategies over model tuning
📍 Evaluate transferability with hidden regions
Dive in: Join the challenge on Dynabench
Explore more: CropHarvest Dataset
|