NASA-IBM AI4Science

NASA's Scientific Visualization Studio

Foundation Models for Science: Open Source Science AI

NASA, IBM, and many other academic institutions and organizations collaborate to develop open-source AI foundation model architectures for scientific applications. By leveraging transformer models pre-trained on petabytes of data using self-supervised masked autoencoders, they aim to make complex science domain data more readily usable for all researchers. The partnership has yielded several open-source foundation models, each trained on NASA datasets:

• Surya Heliophysics Model: The Surya model supports scientists studying space weather, solar flares and active regions, and solar winds. The model was trained on fourteen years of solar imaging data captured by Solar Dynamics Observatory (SDO).

  The Surya Heliophysics Foundational Model would not have been possible to develop without exceptional collaborators. This includes NASA’s Marshall Spaceflight Center (Impact AI), IBM Research, NASA’s Heliophysics Division, NASA’s Goddard Space Flight Center, NASA’s Jet Propulsion Laboratory, National Science Foundation’s National Artificial Intelligence Research Resource (NAIRR), University of Alabama in Huntsville, Search for Extraterrestrial Intelligence Institute, Southwest Research Institute, University of Colorado Boulder.

• NASA-IBM Lunar Foundation Model: This model is the most recent addition to the collection of foundation models for science. Planetary scientists can adapt the foundation model for lunar research tasks such as mapping craters, spotting young volcanic features, and estimating where water ice may exist near the lunar poles. The model architecture relies on high-resolution imagery from NASA's Lunar Reconnaissance Orbiter (LRO) complemented by topographic and gravitational data from NASA's GRAIL mission and JAXA’s SELENE spacecraft.

  The collaboration for NASA-IBM Lunar Foundation Model includes NASA’s Marshall Spaceflight Center (Impact AI), IBM Research, NASA’s Planetary Science Division, NASA’s Ames Research Center, NASA’s Goddard Space Flight Center, University of Alabama in Huntsville, Search for Extraterrestrial Intelligence Institute, University of Maryland (Baltimore County), Howard University, and USRA.

By embracing the principles of open AI and open science, IBM and NASA are actively contributing to the global mission of promoting knowledge sharing and accelerating innovations. We hope that the AI community find our efforts useful and that our models help fuel progress in addressing critical science challenges.

If you have any questions, please visit our discussions page.

Have a unique use case using one of our models? We'd love to hear about your work. Email us at hq-openscience@nasa.gov.