IBM and NASAspecific your Marshall Space Flight Centergoing to to collaborate in the development of foundational Artificial Intelligence models with which they will discover new elements and concepts among the enormous volume of geospatial and Earth scientific data held by NASA. The joint work, which will use IBM’s AI, will for the first time apply foundational AI model technology to NASA’s Earth-observing satellite data.
A foundational model is a type of Artificial Intelligence model that is trained on a very large set of labeled data. They can be used in different tasks, in addition to applying the information collected in a specific situation to a different one. They are mostly found in work being done in natural language processing (NLP) technology, an area in which they have seen a notable takeoff in the last five years. IBM, for its part, is one of the pioneering companies in the use of these models in areas other than language.
This partnership between IBM and NASA is intended to offer researchers on Earth-related topics a simpler system to analyze and extract information from the data they have about her. With IBM’s foundational modeling technology they can develop both their analysis and their discovery. This allows them to advance more quickly in the scientific understanding of the Earth and what happens on it, which leads to better answers to problems related to climate change.
IBM and NASA have plans to develop new technologies to extract information from Earth observations thanks to their joint work. For starters, one project will train an IBM geospatial intelligence model using the Harmonized Landstat Sentinel-2 (HLS) dataset, on a record of the land surface and the land-use changes they capture. satellites in the orbit of the planet.
By analyzing the vast amounts of collected satellite data to identify changes in the geographic footprint caused by events such as natural disasters and crop yields, this foundational model will help researchers to have a critical analysis of systems. Earth’s environments.
The collaboration between both entities is also expected to create a corpus of literature on Earth sciences, which can be used to easily search for information. IBM has created a PLN model trained on some 300,000 earth science journal articles to classify earth science literature and facilitate knowledge discovery.
The model has one of the largest AI workloads trained with Red Hat OpenShift to date. This fully trained model uses PrimeQA, a multi-language question and answer system from IBM, developed using open source. And in addition to the study of Earth sciences, and to understand the effects of climate change and help stop it, it could be integrated into NASA’s scientific data management processes.
Another project that will likely come out of this deal is the development of a foundational weather and climate prediction model based on the MERRA2 dataset. This project is part of the NASA Open Source Initiative.
Raghu Ganti, Principal Investigator at IBMhas pointed out that “foundational models have proven successful in natural language processing, and it is time to expand it into new domains and modalities important to business and society. Applying foundational models to geospatial, sequence of events, time series, and other non-linguistic factors within Earth science data could make enormously valuable insights and information available to a much broader group of researchers, companies and citizens. Ultimately, it could help a larger number of people working on some of our most pressing climate issues.«.
For his part, Rahul Ramachandran, Principal Investigator, NASA Marshal Space Flight Centerhas recalled that «the beauty of foundation models is that they can potentially be used for many downstream applications“, although acknowledging that”the construction of these models cannot be tackled by small teams. Teams in different organizations are needed to bring their different perspectives, resources, and skill sets to the table.«.
