On September 2,According to NASA's official websiteResearchers at NASA's Jet Propulsion Laboratory have used machine learning to develop experimental computer models that are expected to improve the accuracy of detecting rapid enhancement events. The study could help predict the severity of a hurricane within 24 hours, giving people along the way more time to prepare.
Machine learning aids for predicting hurricane intensity
(hurricane Laura went through a rapid strengthening process shortly before landing in Louisiana, with wind speeds jumping to 35 mph or higher in 24 hours. (photo: NASA website)
Hurricane forecasts are divided into two parts: track and intensity. After screening satellite data for many years, the researchers selected several factors: rainfall rate, ice water content and reserved temperature, and added them to the running model of the National Hurricane Center of the United States, and obtained their own prediction through machine learning. NASA says machine learning is better able to analyze complex internal dynamics and determine which factors may cause a sudden increase in hurricane intensity.
this machine learning model is implemented using IBM Watson Studio algorithms. Using the storm data training model from 1998 to 2008, the team used the storm data test model from 2009 to 2014 and compared the test results with the prediction model of the National Hurricane Center in the United States. Fo
NASA's AI attempt
At the AI world government conference in Washington, D.C., on June 26, 2019, NASA data scientist and open innovation program managerBrian Thomas talked about itNASA is using AI / ml to do four types of work: aviation, operations, human capital, and it support.
The last thing NASA wants to see is a wing or model blown up in a wind tunnel, Thomas said, but they are now unable to model using traditional models and techniques to understand the complex structure of the real world (space). It is believed that in aviation, NASA can build better spacecraft and equipment through AI / ml.
On the operational side, Thomas said NASA could use AI / ml technology to assist antenna positioning to maximize access to satellites. In terms of human resources, NSAs plans to analyze job descriptions through text analysis, and then classify them to predict what needs to be done and where employees should be. It support is convenient, including AI / ml to help understand network intrusion and how to avoid intrusion.
Here are some of NASA's specific AI applications, helpers and trial cases.
A paper published on September 1 introduces the MMS algorithm that AI assistants help NASA scientists. Matthew Argall, a space physicist at the University of New Hampshire and lead author of the paper, said the first machine learning task implemented by MMS was NASA's. The mission of MMS is to detect when a spacecraft can cross from the earth's magnetic field to the solar magnetic field, and vice versa.
In the case of no AI assistant intervention, it is necessary to manually retrieve the valuable data in 84 hours within 24 hours, and then select the most appropriate shuttle moment. In the past, 73 trained volunteers took turns on duty to ensure the best data was transmitted to the ground. This is believed to add to the pressure of volunteer service in the busy work schedule of researchers.
As a result, AI short hair mimics how humans read data and identify magnetic insulators according to experience. The difficulty is that most neural networks process data in isolated snapshots, and scientists' observations and predictions are dynamic over time. So by storing the network, seeing past data and adding future data, the research team provides the ability to make current decisions in Beijing, close to scientists,
At present, the new algorithm coincides with human judgment about 70%.
In July 2018,NASA releases newsCimon will become the first batch of intelligent assistants to be used in the space station, and its main task is to handle routine tasks.
CIMON are head-shaped AI robots available on the space station, developed by Airbus and IBM with funding from DLR
(source:DLR German Aerospace Center)
A news report in November 2019 showed that NASA scientists are trying to work with technology companies in Silicon Valley to apply AI algorithms to space science.
NASA astrobiologist at the Goddard Space Flight Center Giada Arney hope that machine learning can help her and her colleagues find the source of life in a pile of data collected by telescopes and observatories,
To help more scientists like Arney achieve their goals, NASA'sFrontier Development LabAn eight week summer program is held every summer, bringing together researchers in the scientific field and data scientists to do research.
This year's research contentThese include: solar physics; planetary science; earth science, such as forecasting future drought images; disaster management, such as lightning and extreme weather forecasting, and mapping flood inundation maps; astrophysics, such as searching for unusual stars and planetary systems; astronaut health, and the establishment of causal inference test-bed.
In 2017, the machine learning program developed by FDL participants can quickly create 3D models of nearby asteroids, so as to accurately estimate their shape, size and rotation speed, which are crucial for detecting and deflecting asteroids that pose a threat to the earth.
In December 2017, NASA said the Kepler space telescope had new discoveries. The latest discovery was made using Google's machine learning technology, using AI to discover the eighth star orbiting the earth.
This year's FDL summer program mentioned earlier has challenges in finding unusual stars and planetary systems:
Recently, NASA's space telescope mission has discovered strange stars whose brightness changes with time. At the same time, each newly discovered solar system challenges our understanding of the evolution of the solar system. But the search space is huge. Can machine learning help?
Recently, NASA has cooperated with Microsoft to create three Python based learning modules, which can teach beginners to explore space using Python and machine learning algorithms, classify space rocks, and predict weather and rocket launch delays.
To sum up, it can be seen that many of NASA AI explorations are designed to help astronomers or data scientists process complex space data. As Thomas said,