NASA partners with IBM to create AI-based models to advance climate science – AI News Update

american space agency Nasa is not only concerned with the exploration of outer space, it is also concerned with helping humanity learn more about planet Earth and the impacts of climate change.

Today, NASA and IBM announced a partnership that will see the development of new artificial intelligence (AI) basic models to help analyze geospatial data from satellites, with the aim of better understanding and acting on climate change. To date, NASA has largely relied on developing its own set of bespoke AI models to address specific use cases. The promise of the foundation model approach is a large language model (LLM) which has been trained on a lot of data that can serve as a more general purpose system that can be customized as needed.

One of the initial goals of the partnership is to train a base model on NASA Harmonized Landsat Sentinel-2 (HLS), which contains petabytes of data collected from space on land use changes on Earth.

In addition to helping improve the state of climate analysis on Earth, IBM hopes the new baseline model it is jointly developing with NASA will have broader applicability and positive impact for use cases. of AI in business.

“What we’re doing with NASA is going to help push innovation from infrastructure and hardware to distributed systems platforms, middleware, and the applications themselves,” Priya Nagpurkar, vice president, hybrid cloud platform and developer productivity at IBM Research, said during a press briefing announcing the partnership. “And that will include advancing AI architectures, and even data management techniques.”

Houston, we got a problem (big data)

To say the least, NASA has plenty of data.

Rahul Ramachandran, principal investigator at NASA’s Marshall Space Flight Center in Huntsville, Alabama, explained at the press conference that NASA actually has the largest collection of Earth observation data. This data was collected to support NASA’s science mission to understand planet Earth as a complex system. The data comes from a variety of instruments and currently includes an archive of 70 petabytes of data. The archive should grow within a few years to reach 250 petabytes.

“Obviously, given the breadth of data we have, we have a big data problem,” Ramachandran said. “Our goal is to make our data discoverable, accessible and usable for broad scientific use in applications around the world.”

Ramachandran added that NASA is always looking for new approaches and technologies that will help streamline the research process, as well as lower the barrier to entry for end users to use the complex scientific data held by NASA. space agency. This is where the development of basic models comes in to more easily take advantage of the data that NASA has collected.

The potential of the base model that NASA is building with IBM could literally change the lives of mankind.

For example, Ramachandran said building a foundation model containing satellite image data could make it easier for someone in a disaster area to identify the extent of the flood, where the model maps. automatically where flooding occurs. Another example might be identifying damage in a hurricane area.

PyTorch and open source AI will also benefit

On the technology side, IBM will make extensive use of a suite of technologies, including Red Hat OpenShift, to run AI training workloads and open-source machine learning frameworks, including TorchPy.

Open source PyTorch machine learning framework launched on Facebook (now known as Meta) and made its own PyTorch Foundation in September 2022. IBM Research has actively contributed to PyTorch, integrating capabilities into the PyTorch version 1.13 framework to help run large workloads on commodity hardware.

Raghu Ganti, Principal Investigator at IBM Research, said PyTorch is a central part of IBM’s AI strategy.

“We rely solely on PyTorch to train all of our base models,” Ganti said.

Ganti added that IBM will continue to contribute to the PyTorch community while continuing to innovate on the technology to create increasingly powerful base models. According to Ganti, the joint effort with NASA to build base models will have multiple applications and broad impact.

“I think it will augment and speed up the scientific process in terms of building and solving specific scientific problems,” he said. “Instead of people having to create their own individual machine learning pipeline, starting with collecting large volumes of training data.”