Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
Space AI NASA

Breakthrough AI Identifies 50 New Planets From Old NASA Data 44

British researchers have identified 50 new planets using artificial intelligence, marking a technological breakthrough in astronomy. CNN reports: Astronomers and computer scientists from the University of Warwick built a machine learning algorithm to dig through old NASA data containing thousands of potential planet candidates. It's not always clear, however, which of these candidates are genuine. When scientists search for exoplanets (planets outside our solar system), they look for dips in light that indicate a planet passing between the telescope and their star. But these dips could also be caused by other factors, like background interference or even errors in the camera. But the new AI can tell the difference.

The research team trained the algorithm by having it go through data collected by NASA's now-retired Kepler Space Telescope, which spent nine years in deep space on a world-hunting mission. Once the algorithm learned to accurately separate real planets from false positives, it was used to analyze old data sets that had not yet been confirmed -- which is where it found the 50 exoplanets. These 50 exoplanets, which orbit around other stars, range in size from as large as Neptune to smaller than Earth, the university said in a news release. Some of their orbits are as long as 200 days, and some as short as a single day. And now that astronomers know the planets are real, they can prioritize them for further observation.
The findings have been published in the journal Monthly Notices of the Royal Astronomical Society.
This discussion has been archived. No new comments can be posted.

Breakthrough AI Identifies 50 New Planets From Old NASA Data

Comments Filter:
  • 7842789424 to go
  • The article talks about how the ML model has been trained to detect these differences and was then set loose on an uncategorized dataset where it came up with these 50 matches. There is no mention that any of the candidates identified by the algorithm have been independently verified to be actual planets, however, they're simply taking the model's prediction at face value. This seems like a bit of a leap, especially as they're developing a whole new and unproven validation model. A more reasonable conclusio

    • We discuss the limitations and caveats of this methodology, and after accounting for possible failure modes newly validate 50 Kepler candidates as planets, sanity checking the validations by confirming them with vespa using up to date stellar information. Concerning discrepancies with vespa arise for many other candidates, which typically resolve in favour of our models.

      From the abstract.

      • by mwvdlee ( 775178 ) on Thursday August 27, 2020 @07:22AM (#60445920) Homepage

        Headline; "Slashdot commenter wrong; scientists not stupid after all".
        In an unprecedented twist, it turns out a slashdot commenter having read the summary for one whole minute did not successfully contradict years of work and decades of accumulated experience of a team of professional scientists.

      • Yeah pretty much that.

        What you'd do would be to give it all the data of stars we already looked at and get it to predict whether us humans found an exoplanet or not. So, clearly it's never going to get better than us, because we're teaching it based on what we can already detect. Since the AI is much less costly to run than a manual analysis, the advantage is that you can set it loose on large amounts of data and it gives you a list of possible stars to look at. You then do the traditional analysis to be su

    • they're simply taking the model's prediction at face value

      Lot of that going around lately.

  • In the future... (Score:5, Interesting)

    by mschaffer ( 97223 ) on Thursday August 27, 2020 @07:09AM (#60445898)

    In the future astronomers will be simply combing through old data from NASA and other sources because the only thing visible from Earth will be Elon's Starlink.

    • Re:In the future... (Score:5, Informative)

      by drinkypoo ( 153816 ) <drink@hyperlogos.org> on Thursday August 27, 2020 @08:48AM (#60446186) Homepage Journal

      They'll be looking at new data from NASA and other sources because Elon's SpaceX will have made launches cheap enough that we'll have a plethora of new space telescopes.

      • Could be. Can you imagine self-adjusting fleets of mirrors around the Earth all focusing on a receiver in geosynchronous orbit?

      • by 4im ( 181450 )

        They'll be looking at new data from NASA and other sources because Elon's SpaceX will have made launches cheap enough that we'll have a plethora of new space telescopes.

        Wishful thinking. Also, consider the work done by backyard "amateur" astronomers... it's far from negligible.

        • Elon's SpaceX will have made launches cheap enough that we'll have a plethora of new space telescopes.

          Wishful thinking.

          Yeah, I suppose it's too much to expect that NASA will be able to take advantage of the situation to do more science when its primary mission is to secure pork. Still, one can hope.

          Also, consider the work done by backyard "amateur" astronomers... it's far from negligible.

          Sure, it has meaning. But so does providing internet access to the masses. As one of the many people who have shitty internet access now, I have little sympathy for the masses who impeded my attempts to get better access in the past, and who are still doing it now. When I explain to them that every major telco is a criminal consp

    • Starlink is in a fairly low orbit. Terrestrial telescopes may have a lot of noise, but orbitals should be fine.

      • I'll bet there's some lobbying going on from Boeing et. al., complaining that Starlink is polluting the skies before they get to.

  • We live on earth and there is only one around for at least 100 generations. Can we just fix our situation? Which is infinitely easier than going at Z million light-years from earth and can be done during our lifetime. thank you
    • Can we not strawman arguments into irrelevant rants? The article is about how AI used to reanalyze data detecting things that humans have missed. No one has started a GoFundMe page to build rockets to those planets. That AI could be used for other purposes here on Earth.
      • That AI could be used for other purposes here on Earth.
        No, it could not.
        For starters: it is not an AI but an artificial neuron network.
        Secondly: it is trained for a certain purpose. So, you have to train it - if you can - for another purpose, which might take years.

        • Let me understand your statement: "No, it can’t be reused and it will take years to be reused." Did I state that correctly? First of all your assumption is that only the current system in its exact form can only ever be used. That in itself is a bad assumption. Parts of it could be reused for other roles. Second, the fact it took years to hone it to this specific purpose does not mean it require years to structure to another purpose. The researchers could have learned on how to quickly change the syst
          • Perhaps you should read up how an artificial neural network works.
            No, it can not be reuses.

            If they have a fancy idea how to construct that particular network which makes it usefull for particular cases of "recognizing patterns".

            Then it can be "reused" as in _trained_ for different subjects, as in chemistry or biology aiming at "similar patterns".

            The original AI/NN can not be re used at all for anything else. How would you train it with, let's say chemical reactions, and then let it look at planet finding pi

            • Perhaps you should read up how an artificial neural network works. No, it can not be reuses.

              1) Perhaps you should read what you said. We can both scroll up and read what you said: "For starters: it is not an AI but an artificial neuron network. Secondly: it is trained for a certain purpose. So, you have to train it - if you can - for another purpose, which might take years."

              You contradicted yourself in saying it cannot be done AND it will take years for it to be done. You said this. Scroll up.

              2) I said: " That AI could be used for other purposes here on Earth." It's not a long statement. I never s

              • I did not contradict myself.

                You seem not to grasp what I say.

                A neural network trained to analyze pictures, to regognize cats, e.g. can be trained to analyze other pictures.

                It most likely can not be trained to analyze variations of the shares market. You most likely need a different architecture of neural network.

                Hence: AI/NNs of one kind rarely can be "reused". The minimum is clearing it and retraining it.

                Seriously, read some wikipedia about it.

                • Hence: AI/NNs of one kind rarely can be "reused". The minimum is clearing it and retraining it. Seriously, read some wikipedia about it.

                  Whatever your games you want to play with nomenclature, you don't seem to want to admit that it indeed can be reused. Are they destroying the whole network and burying it in a yard never to be used again? Oh they'll use the same hardware, the same techniques, etc. In other words: REUSE. Please read a dictionary and learn what the word reuse means.

                  • The one who is trying to play games is you.

                    Again: you can not reuse a NN for anything else it is not trained for.

                    Grasp it or don't grasp it, up to you.

                    • Again: you can not reuse a NN for anything else it is not trained for.

                      So they just take the NN out back, shoot it and bury it in the yard. Every time the researchers develop a new NN, they burn all the work they've previously done and start from scratch? Or do they REUSE research including parts or all of the old NN. Your statement is like saying no computer is built from scratch and not based on any other computer.

                    • Ofc. the reuse the research.
                      But that was not the topic.

                    • So they destroyed all the hardware too? Every NN is built on completely brand new hardware? You just admitted they reused the research that built the NN. Some of that would be the software code? Isn’t that part of the NN? And again I have to remind you that I never said the exact original NN was used. I never said that and clarified that point long ago.
                    • An NN is software. Not hardware, besides the fact that we now have special "CPUs" tailored for NNs just like GPUs are tailored for graphics. Actually you could abuse GPUs to build NNs, the math is similar. But it would still be just software (and data) using a special purpose processor.

                    • An NN is software. Not hardware, besides the fact that we now have special "CPUs" tailored for NNs just like GPUs are tailored for graphics. Actually you could abuse GPUs to build NNs, the math is similar. But it would still be just software (and data) using a special purpose processor.

                      A NN is software and sometimes specialized hardware?

                      Again, my point is do they throw out everything if they wanted to build a new NN? No, they may REUSE the hardware especially if it is specialized hardware. They may REUSE the code. Do you see a pattern here?

                    • No, they do not reuses the code ... seriously, stop discussing.
                      You ran into a maze when you started with your first post: "I hope they can reuses it for something else".

                      The only software, you can reuse .if you use a pure software one - is a simple library that does nothing out of it self.

                      The crux of an NN its its architecture. For simpletons like you: how many layers does it have, how are they connected? And: the data used to train it. And: the algorithm(s) used to correct its errors.

                      If one invents a new NN

                    • No, they do not reuses the code ... seriously, stop discussing.

                      So they throw out all the code and start from scratch with every NN then? So what do they save on that specialized hardware when they have to rewrite code all the time? I mean if they never reuse the code, then what is saved on the hardware? Is that where they keep their porn collection?

                      The only software, you can reuse .if you use a pure software one - is a simple library that does nothing out of it self.

                      And where did that library come from? Is that like a Python plugin I can just install on my Android phone then? The University of Warwick researchers sound like a bunch of lazy chaps then. Download a library and run it on da

  • You make yourself look stupid and you make me feel stupid for having read what you wrote.

No skis take rocks like rental skis!

Working...