SoylentNews is people
SoylentNews
NASA will be hosting a somewhat unusual press conference on Thursday (NASA will host a media teleconference at 1 p.m. EST Thursday, Dec. 14) to announce the latest find from its planet-hunting Kepler Space Telescope. Kepler has found many hundreds of planets beyond our solar system over the years, but this week's announcement will be different because Google will be sharing in the science spotlight.
"The discovery was made by researchers using machine learning from Google," reads a release from the space agency, adding that the breakthrough "demonstrates new ways of analyzing Kepler data."
Exactly what has been discovered won't be revealed until Thursday, but with Kepler there's always a good chance that some new distant planets will be part of the reveal. Expect to hear something about a new era of planet-hunting assisted by artificial intelligence: That would be my guess for Thursday. We'll just have to wait and see if Google's A.I. is also helping to detect signs of alien life on the numerous worlds beyond our solar system as well.
https://www.cnet.com/news/google-nasa-kepler-artificial-intelligence-machine-learning-planets/
(Score: 5, Interesting) by bzipitidoo on Wednesday December 13 2017, @12:26PM (2 children)
But in this case, the answer is not that hard to figure out, even if the article hadn't all but revealed it anyway. Google is making a big splash in AI, with these recent triumphs in the classic strategy games Go and chess. Now, what uses are there for this superhuman AI that is so good at learning and pattern recognition?
If you know a little about astronomy, you know that there are petabytes of data archived from years of sky surveys which no one has ever examined because there aren't enough astronomers to look at everything. But employing a simpleminded automatic scan to dig through all that data for interesting objects isn't practical either. The images are full of noise and artifacts. Simple scanning isn't up to that job, need people to judge whether a few odd looking pixels is just an artifact or an error, or a known object, or something new to us. A similar area is Optical Character Recognition. OCR is one of those exasperating problems that just doesn't seem like it can be so hard for computers to do, but it is. OCR algorithms are poor at reading printed text, and make all kinds of stupid, brainless mistakes a person would never make.
Enter Google's hot new AI, AlphaGo Zero. It probably can quickly learn to pick out and identify all the stars in typical, noisy photos of the night sky. I have no doubt one trained instance of Google's neural network AI can examine photos by the thousands per hour, and present human astronomers with the short list of all unknown objects in those photos, and be almost 100% correct.
(Score: 3, Informative) by takyon on Wednesday December 13 2017, @01:05PM
https://en.wikipedia.org/wiki/Kepler_(spacecraft)#Confirming_planet_candidates [wikipedia.org]
If all they are presenting is Kepler data (nothing from other telescopes like Spitzer), they are likely to add to the list of exoplanet candidates, but they (Kepler team) can also confirm the exoplanets using astrometry.
We can expect the results to be about Earth-sized exoplanets in habitable zones.
Kepler is good at determining the radius and orbital period of an exoplanet, but not necessarily the mass. But it CAN be done. For example:
Measuring the Mass of a Mars-size Exoplanet [nasa.gov]
The mass of the Mars-sized exoplanet Kepler-138 b from transit timing [nature.com]
By finding tiny changes in transit timing caused by systems with multiple exoplanets, mass, and thus density, can be constrained.
Kepler data + machine learning can get you almost everything you need to describe the exoplanet, but not atmospheric composition AFAIK.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 4, Funny) by bob_super on Wednesday December 13 2017, @06:08PM
Silly people. I learnt long time ago that the way to find out whether a celestial object exist is to check the Jedi database.