IBM Corp. and Nasa have unveiled an advanced artificial intelligence (AI) Model aimed at accelerating the analysis of satellite data. The model is now available on Hugging Face, a platform similar to GitHub for sharing open-source neural networks. This release marks the next phase of IBM’s collaboration with NASA and their plans to expand the AI’s use cases. The two organizations will partner with Clark University in Worcester, Massachusetts.
The primary objective of this AI model is to help researchers identify areas in the continental U.S. that are susceptible to flooding and wildfires. IBM claims that their model can analyze geospatial data up to four times faster than existing state-of-the-art neural networks while requiring less data for training.
This AI model, based on the Transformer architecture, serves as a foundation model capable of performing various advanced computing tasks. Transformer models like this one excel at utilizing contextual information to make more accurate decisions. It is worth noting that the Transformer architecture is also used in OpenAI LP’s latest large language model, GPT-4.
IBM and NASA trained their model on a geospatial dataset called Harmonized Landsat Sentinel-2, which comprises images of the Earth’s surface captured by NASA’s Landsat-8 satellite and measurements from the European Space Agency’s Sentinel-2 satellite constellation.
IBM utilized its internally-developed Vela supercomputer, powered by Nvidia Corp.’s A100 series of data center graphics cards, to train the AI model. The supercomputer incorporates IBM’s virtualization software, which reduces processing power while providing ease of use. Despite the reduction in performance, IBM claims to have achieved the “lowest overhead in the industry.”
While the model was optimized for detecting areas at risk of flooding and wildfires, IBM and NASA believe it can be adaptable to other use cases as well. Potential applications include tracking deforestation, monitoring carbon emissions, and forecasting crop yields.
IBM plans to further expand the capabilities of this AI model through collaboration with NASA and Clark University. Additionally, the company intends to develop a commercial version of the model for its Watsonx product suite, which offers software tools and prepackaged neural networks optimized for various use cases. The suite is powered by Red Hat OpenShift AI, a component of IBM’s machine learning portfolio designed specifically for AI workloads.
The post IBM and NASA Release Advanced AI Model to Analyze Satellite Data appeared first on TS2 SPACE.
