NASA Applies AI Technology to Problems in Space Science

2019-11-24

Machine learning is a type of AI that enables computers to learn from data using a series of algorithms and other tools, allowing them to make predictions or classify target objects faster and more accurately than humans. As a result, machine learning is widely used in facial recognition or recommendation algorithms. However, some scientists believe that the application of machine learning goes far beyond these areas, even extending to extraterrestrial domains.


Can computer algorithms trained to ensure the safe driving of autonomous vehicles help identify nearby asteroids or discover life in the universe? This is a question that scientists at NASA have been pondering, as they have been attempting to apply advanced computer algorithms to problems in space science to find out. To assist scientists in building state-of-the-art research tools, NASA's Development Laboratory brings together technology and space researchers every summer to collaborate and develop computer code. This four-year program is a partnership between the SETI Institute and NASA's Ames Research Center, both located in Silicon Valley, where incubators bring talent together to accelerate the development of breakthrough technologies.


In the previous research process at NASA, the laboratory paired doctoral students in science and computer engineering with experts from the space agency, academia, and some of the world's largest technology companies. This collaboration contributed various combinations of hardware, algorithms, supercomputing resources, funding, facilities, and subject matter experts. All AI technologies developed by the laboratory will be publicly available internally, with some already helping identify asteroids, search for planets, and predict extreme solar radiation events.


To this end, scientists Arney and Domagal-Goldman from NASA's research team have developed a new technology with the support of Google AI, deploying Bayesian neural networks in space telescopes. The team wanted to know if Bayesian neural networks could perform better.


Arney said, "We immediately found that Bayesian neural networks have better accuracy in identifying the abundance of various molecules in the atmosphere of WASP-12b compared to random forests."


But besides better accuracy, Bayesian technology also provides something equally important: it can tell scientists the certainty of predictions. Domagal-Goldman said, "Where data is insufficient to provide true accurate results, the model better predicts incorrect results, which is crucial for our confidence in these predictions."


In addition, other technologies developed by the team have been successfully applied. For example, researchers previously developed a machine learning program that can quickly create 3D models of nearby asteroids, accurately estimating their shape, size, and rotation speed. This information is crucial for NASA's work in detecting and deflecting asteroid threats from Earth.


The potential of this technology is not lost on anyone. Damon, the project manager of NASA's technology program, envisions integrating these virtual tools into spacecraft in the future, making missions lighter, simpler, and therefore more cost-effective. Domagal-Goldman and Arney envision future missions to exoplanets, where AI technology embedded in spacecraft is intelligent enough to make real-time scientific decisions, saving a lot of time that would otherwise be spent communicating with Earth scientists.

RECENT AI NEWS

RECENT AI TOOLS