Observation, simulation, and AI join forces to reveal a clear universe


Research News


Japanese astronomers have developed a new artificial intelligence (AI) technique to remove noise in astronomical data due to random variations in galaxy shapes. After extensive training and testing on large mock data created by supercomputer simulations, they then applied this new tool to actual data from Japan’s Subaru Telescope and found that the mass distribution derived from using this method is consistent with the currently accepted models of the Universe. This is a powerful new tool for analyzing big data from current and planned astronomy surveys.

Wide area survey data can be used to study the large-scale structure of the Universe through measurements of gravitational lensing patterns. In gravitational lensing, the gravity of a foreground object, like a cluster of galaxies, can distort the image of a background object, such as a more distant galaxy. Some examples of gravitational lensing are obvious, such as the “Eye of Horus”. The large-scale structure, consisting mostly of mysterious “dark” matter, can distort the shapes of distant galaxies as well, but the expected lensing effect is subtle. Averaging over many galaxies in an area is required to create a map of foreground dark matter distributions.

But this technique of looking at many galaxy images runs into a problem; some galaxies are just innately a little funny looking. It is difficult to distinguish between a galaxy image distorted by gravitational lensing and a galaxy that is actually distorted. This is referred to as shape noise and is one of the limiting factors in research studying the large-scale structure of the Universe.

To compensate for shape noise, a team of Japanese astronomers first used ATERUI II, the world’s most powerful supercomputer dedicated to astronomy, to generate 25,000 mock galaxy catalogs based on real data from the Subaru Telescope. They then added realist noise to these perfectly known artificial data sets, and trained an AI to statistically recover the lensing dark matter from the mock data.

After training, the AI was able to recover previously unobservable fine details, helping to improve our understanding of the cosmic dark matter. Then using this AI on real data covering 21 square degrees of the sky, the team found a distribution of foreground mass consistent with the standard cosmological model.

“This research shows the benefits of combining different types of research: observations, simulations, and AI data analysis.” comments Masato Shirasaki, the leader of the team, “In this era of big data, we need to step across traditional boundaries between specialties and use all available tools to understand the data. If we can do this, it will open new fields in astronomy and other sciences.”


These results appeared as Shirasaki et al. “Noise reduction for weak lensing mass mapping: an application of generative adversarial networks to Subaru Hyper Suprime-Cam first-year data” in the June 2021 issue of Monthly Notices of the Royal Astronomical Society.

From EurekAlert

3.9 9 votes
Article Rating
Newest Most Voted
Inline Feedbacks
View all comments
July 3, 2021 10:25 pm

So the AI is trained based on our present understanding of the universe. Strangely the AI corrected view is consistent with the present cosmological model. Strange! /sark

Rich Davis
Reply to  Lars Silen
July 4, 2021 6:29 am

My first thought as well. How do you add realist (sic) noise (presumably that’s “realistic”) to your synthetic galaxy data, except by assuming that the noise is the difference between observations and your “currently-accepted” model of the universe? Then you train your AI to remove “noise” and shockingly, it adjusts the observations to look exactly like we expected the universe to appear!

My second thought was to wonder who’s presenting this rubbish. No surprise there—our EurekAlert! friends.

Reply to  Rich Davis
July 4, 2021 2:43 pm

Have the climate modelers found a new job?

Reply to  Lars Silen
July 4, 2021 7:55 am

There does seem to be some circular processes here…

July 3, 2021 10:32 pm

Why didn’t they ask a fortune teller?
Would be much cheaper.

Harry Passfield
Reply to  Alex
July 4, 2021 5:00 am

Damien Hirst seems to have been rather prescient. /s

Rich Davis
Reply to  Harry Passfield
July 4, 2021 9:00 am

That’s what I love about this site. Now I know who Damien Hirst is. It does look like he provided the “realist noise”.

Peta of Newark
July 3, 2021 11:34 pm

Quote:new artificial intelligence (AI) technique to remove noise”

They ought to be very careful how they go, someone may use that very thing on them.
(Please please please ‘someone, use it on Climate Science first)

Here’s a thing to think about, although those familiar with Carnot will know all about it.

Q: Why is the universe cooling?
A: Because there isn’t anywhere, or anything, within it at a temperature of zero Kelvin

This applies of course to the GHGE, where everyone asserts as Rud did recently, that Green Gases absorb energy then re-emit it in all directions.

Yes and No

As ststed by Carnot and the lack of a Zero Object, all those interactions are less than 100% efficient.
IOW: The energy re-emitted is NOT the same energy as was absorbed. It is why power stations have cooling towers and why your car engine has both a tail-pipe and a radiator (so called even tho it is in fact a conductor/convector)

Someone even wrote an essay on here recently about ‘Spent Energy’. Patently many, Rud included, either didn’t read it or let it go over their heads.

That is why the universe is cooling.
If a Zero Kelvin object did exist, ever since the very instant of the Big Bang (if such a thing did happen), all interactions between Energy and Matter would have been 100% efficient.
Thus, what is stated to happen inside the GHGE would be true. i.e. Gases absorb then re-emit equally in all directions.

But, that 100% efficiency of energy/matter interaction would mean that the Big Bang Universe would not ever cool down. That it would remain forever at whatever temperature it was at its moment of creation.
Exactly because there is nowhere in the universe with a temperature of zero Kelvin to act as an exhaust (or Tc for those familiar with Carnot’s equation) for energy/matter interactions – each and every one of them being a miniature Carnot Heat Engine

Ain’t that the craziest thing?
Most folks would expect the exact opposite to occur – that the zero Kelvin object would suck all the energy out of the universe – instead of conserving it at ever increasing wavelegths and lower temps. It creates/defines both the 1st and 2nd laws

It don’t necessarily mean that the GHGE is complete garbage, just that every explanation of it is complete garbage.

So. Is the Universe cooling down or is the GHGE correct?
You Can Not Have Both = True

Makes you wonder though doesn’t it?
Maybe the GHGE is just ordinary bog-standard Entropy in action, that there is in fact nothing magical going on in there at all
Maybe Big Don’s T-Total instinct was/is correct. Climate Change IS a hoax
We all have that instinct, try using it sometimes…….

Reply to  Peta of Newark
July 4, 2021 12:45 am

Ok, but universe is not in an Ice Age.

Reply to  gbaikie
July 4, 2021 12:11 pm

Dark Matter might be simply ice, and the universe, especially around galaxies, is full of ice. Maybe the universe is in a quasi Ice Age. My worst nightmare…I am dreading winter already and still in a heatwave out here in the west.

John Endicott
Reply to  Earthling2
July 7, 2021 4:57 am

If it’s Dark Matter than surely it’s in a Dark Age instead. (I know, Don’t call you Shirley).

July 4, 2021 4:42 am

What if the universe we can see is actually a black hole?  But we know that there are other black holes inside our black hole, so then the multiverse is a Russian doll set of nested black holes. Then what is the fate of a photon emitted in a random direction inside our black hole? It will eventually encounter the Schwarzschild radius (SR) from the inside, a region where time stops and nothing passes through. Approaching from the outside it would stop, go infrared and look black, hence the name. But these black holes have axial jets, so energy does not accumulate indefinitely. Before reaching the SR it turns and heads for one of the poles and exits via a polar jet which can be observed. But what about our black hole? Approaching the SR from inside it would go infrared and add to the energy of the SR. Does this energy build indefinately or does something eventually happen? Like break out of our universe and join the polar jet of the superior universe via a polar “worm hole?” I think Stephen Hawking has theorized that black holes eventually evaporate, but his mechanism is a little different. I will now exit this train of thought and go get breakfast.

Reply to  pochas94
July 4, 2021 6:28 am

What if the universe we can see is actually a black hole?

Geodesic incompleteness would happen. Every photon would loop back on itself.

Mike McMillan
Reply to  Hatter Eggburn
July 4, 2021 8:05 pm

I have those days sometimes.

July 4, 2021 4:53 am

Deep learning can also resolve the filamentous architecture of the local cosmic web of galaxies, again confirming the existence (in an indirect, circumstantial way) of dark matter:


July 4, 2021 5:20 am

The very furthest distance Hubble can “see” is about 15 billion light years out, and beyond that is the Red Blob stuff.
The real stuff outdoes modeling and AI interps every time. AI can be wrong. It is based on limited information and best guessing, not real-world stuff. Modeling is best-guessing.

Tom Abbott
Reply to  Sara
July 4, 2021 5:49 am

Musk needs to mount a rescue mission to the Hubble Space Telescope if NASA can’t get its computer functioning properly again. NASA is having problems figuring out what the problem is with Hubble’s computers.

It would be handy to have an orbital transfer vehicle available right about now.

Rich Davis
Reply to  Tom Abbott
July 4, 2021 6:42 am

Maybe a souped up Subaru Outback to be called the Subaru Way Outback. I imagine that’s how the Japanese are servicing the Subaru Telescope?

Reply to  Rich Davis
July 6, 2021 6:19 am

On that line, how ’bout Peabody and Sherman’s Wayback machine?

July 4, 2021 9:13 am

In measurements you have to eliminate noise or other influences to get a proper value. Their goal is to do that.

Rich Davis
Reply to  Olen
July 4, 2021 12:14 pm

And how do you detect the difference between gravitational lensing and a random variation that is real? They are not talking about a randomly fluctuating signal. They are talking about a stable image that is distorted by gravitational lensing. Not like a transient “twinkling” caused by variation in our atmosphere.

July 4, 2021 12:54 pm

Much of what we can see in space/time has to be an illusion, at least to us from our vantage point and to anything anywhere else in the universe looking towards us. Given the variable distances involved to everything and the fact the universe appears to be expanding due to dark energy, everything is an illusion, not to mention the gravitational lensing they mention they are trying to use AI to figure out what things really look like. What happens when galaxies 10 billion light years away have been gravitationally lensed near infinite times between us and those objects so far away? How would we know what to correct for? 

What happens to all the tens of trillions of tons of plasma that gets blown off every star all the time? Or all the photons that have no mass, but do have momentum. Light is composed of photons, which have no mass, so therefore light has no mass and can’t weigh anything. Maybe not so fast. Because photons have energy — and, as Einstein taught us, energy is equal to the mass of a body, multiplied by the speed of light squared. So what is the ultimate fate of a photon? Dark energy in transit to expanding the universe? 

Or all the solar winds that are blowing off every star in the universe? The solar wind is a stream of energetic particles ejected by the Sun. These include electrons and protons from hydrogen, along with atomic nuclei like helium, otherwise known as alpha particles. There are also traces of ‘heavy ions’ and atomic nuclei of carbon, nitrogen, oxygen, neon and magnesium. Maybe explains the Oort Cloud, which lies far beyond Pluto and the most distant edges of the Kuiper Belt. While the planets of our solar system orbit in a flat plane, the Oort Cloud is believed to be a giant spherical shell surrounding the Sun, planets and Kuiper Belt Objects. Since the solar wind is matter and streaming continuously 360 in 3D, possibly winds up as ice and dust surrounding all the stars. Dark matter? 

Maybe dark matter and dark energy is so simple, it is just staring us right in the face. Fun to speculate, and a great article to think about on a hot lazy Sunday July 4th. Happy 245th Birthday America! Live long and prosper.

July 4, 2021 2:05 pm

Masato Shirasaki, the leader of the team: “In this era of big data, we need to step across traditional boundaries between specialties and use all available tools to understand the data. If we can do this, it will open new fields in astronomy and other sciences.”

How applicable might this technique be to the enhancement of medical imaging?

July 4, 2021 9:58 pm

A friendly reminder that “AI” is just another term for “software”.

BTW whatever happened to multi-sampling to increase S:N? I would not trust software to discover previously “hidden” details.

Reply to  Karim Ghantous
July 5, 2021 11:00 am

If the neural nets of the AI software are properly taught, (this does involve some experimentation and human feedback in its development), then one’s trust in the results is comparatively increased.

John Endicott
Reply to  noaaprogrammer
July 7, 2021 5:04 am

define “properly taught”. And the problem is the human element. Humans have a tendency to program in their biases into the software they create (sometimes inadvertently, sometimes deliberately.) You can only trust the software as far as you are willing to trust it’s creators.

July 5, 2021 8:34 pm

Is there such a thing as Artificial Intelligence? When a computer invents a new technology, all by itself, then I will believe in Artificial Intelligence. The first definition of AI which I read was a computer program which could have a conversation indistinguishable from a human-human interaction. What is the latest definition?

The definition of a Synthesis: “The combination of ideas to form a theory or system.”

Yes, please show me an Artificial one…

Reply to  Michael Moon
July 6, 2021 5:04 pm

I’ve seen some pretty cool card tricks that seemed quite impossible, but the magician was an illusionist performer, he/she was not using actual magic just great theatrics to fool me.

A conversational Ai is the same thing, an illusion of intellect designed only to fool us and everyone realizes actual machine intellect is not going to be possible for reasons so obvious the average bright-minded child will grasp its limits.

But con-men need to make a crust by fooling us with illusions.

John Endicott
July 7, 2021 4:59 am

Gee, a computer programed with our views of the universe see things as being consistent with our views of the universe. Who would have ever guessed?

Last edited 20 days ago by John Endicott
Mark - Helsinki
July 8, 2021 11:04 am

Great, data processing will fix the CBR which is, purely a result of data processing.

Cosmic background radiation map is a result of data processing, not cosmic background radiation for the most part.

There is no way to filter out the intensity of total signal into individual signal, as per tenets of signal detection, when the noise is 2000 times greater than the signals you look for, you need to at least be able to control the signal to tease it out, or have a complete understanding of that signal you haven’t detected yet:D

%d bloggers like this: