[ti:Machine Learning Helps NASA Confirm 301 New Exoplanets] [by:www.51voa.com] [00:00.00]更多听力请访问51VOA.COM [00:00.04]The American space agency NASA [00:03.64]says it has used a new technology method [00:08.44]to help confirm the existence of 301 new exoplanets. [00:17.80]Exoplanets are planets that orbit stars other than the sun. [00:24.20]Before the latest discoveries, NASA had confirmed [00:29.84]the existence of more than 4,569 such planets. [00:37.24]Thousands of other "candidate" exoplanets have been identified. [00:43.72]But these require additional study. [00:48.16]Exoplanets are difficult for telescopes to identify. [00:53.88]One reason is that the bright light of the stars they orbit can hide them. [01:00.76]The search process can involve looking for decreases [01:06.52]in the light level of stars. [01:09.36]Such drops could be caused by a planet passing in front of a star. [01:17.20]NASA has used two space telescopes [01:21.68]to confirm thousands of exoplanets. [01:25.60]The Kepler space telescope was launched in 2009 [01:31.52]and operated until October 2018. [01:36.12]At that time, NASA announced it was retiring Kepler [01:41.60]because the spacecraft had "run out of fuel needed [01:46.32]for further science operations." [01:50.16]The other space telescope is called [01:53.48]the Transiting Exoplanet Survey Satellite, or TESS. [01:59.64]NASA launched TESS in April 2018 [02:04.64]to build on Kepler's observations. [02:08.48]TESS continues to operate today. [02:13.08]NASA's confirmations of the 301 new exoplanets [02:19.16]were based on data collected by the Kepler space telescope. [02:25.00]The data was processed [02:27.88]through a machine learning system called ExoMiner. [02:33.64]Machine learning systems are a form [02:37.20]of artificial intelligence (AI). [02:40.08]They are trained to learn a task over time [02:44.52]by being fed huge amounts of data. [02:49.28]In this case, NASA said it used the machine learning method [02:54.76]to examine existing data to identify real exoplanets [03:00.84]from so-called "imposters." [03:04.76]ExoMiner is powered by data [03:08.16]gathered from past efforts to confirm [03:11.88]or rule out possible exoplanets. [03:15.96]The system was designed to use the same methods [03:20.76]that human experts use to confirm new exoplanets. [03:26.36]NASA said the system provides much-needed assistance [03:32.28]to scientists who are expertly trained [03:35.76]to confirm the existence of such planets. [03:40.36]The agency's space telescopes [03:43.80]collect data on thousands of stars. [03:47.72]It is a huge effort for humans to examine so many stars. [03:53.88]ExoMiner is designed to ease that load [03:58.56]and improve the accuracy of identifying new exoplanets. [04:05.60]Jon Jenkins is an exoplanet scientist [04:10.20]at NASA's Ames Research Center in California. [04:16.28]He said in a statement that ExoMiner offers big improvements [04:22.44]over other machine learning programs [04:25.60]used to identify exoplanets in the past. [04:30.88]The main reason for this, Jenkins said, [04:35.32]is that the new system permits scientists [04:39.48]to easily confirm ExoMiner's findings. [04:44.88]"There is no mystery as to why it decides [04:49.00]something is a planet or not," he said. [04:52.88]"We can easily explain which features in the data [04:58.28]lead ExoMiner to reject or confirm a planet." [05:03.68]The machine learning system was developed and tested [05:08.76]by NASA researchers and the team's international partners. [05:14.32]It was described by a paper [05:18.00]published in the Astrophysical Journal. [05:21.32]The paper explains that ExoMiner discovered [05:26.80]the 301 exoplanets from a list of candidates [05:31.92]based on data from the Kepler space telescope. [05:37.48]They had been identified and declared as possible exoplanets [05:43.44]by scientists at the Kepler Science Operations Center. [05:48.60]But NASA said no human researchers [05:53.20]had been able to confirm them. [05:56.08]"When ExoMiner says something is a planet, [06:00.64]you can be sure it's a planet," said Hamed Valizadegan. [06:06.40]He is the ExoMiner project lead [06:10.56]and oversees machine learning operations [06:14.68]at the Universities Space Research Association [06:19.44]at the Ames center. [06:21.72]Valizadegan added that the system [06:25.68]is "in some ways more reliable" [06:29.12]than both existing machine learning methods [06:32.88]as well as human experts. [06:36.56]He said one reason for this is that ExoMiner [06:41.28]is free of "biases" that can affect human identification operations. [06:48.76]The NASA team said it plans to build [06:52.68]on ExoMiner's success by expanding the system. [06:58.64]The goal would be to include data from TESS [07:02.64]and future telescopes that aim to discover new exoplanets. [07:09.52]I'm Bryan Lynn. 更多听力请访问51VOA.COM