The first AI universe sim is fast and accurate—and its creators don't know how it works

The first AI universe sim is fast and accurate -- and its creators don't know how it works
A comparison of the accuracy of two models of the universe. The new model (left), dubbed D3M, is both faster and more accurate than an existing method (right) called second-order perturbation theory, or 2LPT. The colors represent the average displacement error in millions of light-years for each point in the grid relative to a high-accuracy (though much slower) model. Credit: S. He et al./Proceedings of the National Academy of Sciences 2019

For the first time, astrophysicists have used artificial intelligence techniques to generate complex 3-D simulations of the universe. The results are so fast, accurate and robust that even the creators aren't sure how it all works.

"We can run these simulations in a few milliseconds, while other 'fast' simulations take a couple of minutes," says study co-author Shirley Ho, a group leader at the Flatiron Institute's Center for Computational Astrophysics in New York City and an adjunct professor at Carnegie Mellon University. "Not only that, but we're much more accurate."

The speed and accuracy of the project, called the Deep Density Displacement Model, or D3M for short, wasn't the biggest surprise to the researchers. The real shock was that D3M could accurately simulate how the universe would look if certain parameters were tweaked—such as how much of the cosmos is dark matter—even though the model had never received any training data where those parameters varied.

"It's like teaching image recognition software with lots of pictures of cats and dogs, but then it's able to recognize elephants," Ho explains. "Nobody knows how it does this, and it's a great mystery to be solved."

Ho and her colleagues present D3M June 24 in the Proceedings of the National Academy of Sciences. The study was led by Siyu He, a Flatiron Institute research analyst.

Ho and He worked in collaboration with Yin Li of the Berkeley Center for Cosmological Physics at the University of California, Berkeley, and the Kavli Institute for the Physics and Mathematics of the Universe near Tokyo; Yu Feng of the Berkeley Center for Cosmological Physics; Wei Chen of the Flatiron Institute; Siamak Ravanbakhsh of the University of British Columbia in Vancouver; and Barnabás Póczos of Carnegie Mellon University.

Computer simulations like those made by D3M have become essential to theoretical astrophysics. Scientists want to know how the cosmos might evolve under various scenarios, such as if the dark energy pulling the universe apart varied over time. Such studies require running thousands of simulations, making a lightning-fast and highly accurate computer model one of the major objectives of modern astrophysics.

D3M models how gravity shapes the universe. The researchers opted to focus on gravity alone because it is by far the most important force when it comes to the large-scale evolution of the cosmos.

The most accurate universe simulations calculate how gravity shifts each of billions of individual particles over the entire age of the universe. That level of accuracy takes time, requiring around 300 computation hours for one . Faster methods can finish the same simulations in about two minutes, but the shortcuts required result in lower accuracy.

Ho, He and their colleagues honed the that powers D3M by feeding it 8,000 different simulations from one of the highest-accuracy models available. Neural networks take training data and run calculations on the information; researchers then compare the resulting outcome with the expected outcome. With further training, adapt over time to yield faster and more accurate results.

After training D3M, the researchers ran simulations of a box-shaped 600 million light-years across and compared the results to those of the slow and fast models. Whereas the slow-but-accurate approach took hundreds of hours of computation time per simulation and the existing fast method took a couple of minutes, D3M could complete a simulation in just 30 milliseconds.

D3M also churned out accurate results. When compared with the high-accuracy model, D3M had a relative error of 2.8 percent. Using the same comparison, the existing fast model had a relative error of 9.3 percent.

D3M's remarkable ability to handle parameter variations not found in its makes it an especially useful and flexible tool, Ho says. In addition to modeling other forces, such as hydrodynamics, Ho's team hopes to learn more about how the model works under the hood. Doing so could yield benefits for the advancement of artificial intelligence and machine learning, Ho says.

"We can be an interesting playground for a machine learner to use to see why this model extrapolates so well, why it extrapolates to elephants instead of just recognizing cats and dogs," she says. "It's a two-way street between science and deep learning."


Explore further

CosmoGAN: Training a neural network to study dark matter

More information: Siyu He et al, Learning to predict the cosmological structure formation, Proceedings of the National Academy of Sciences (2019). DOI: 10.1073/pnas.1821458116
Provided by Simons Foundation
Citation: The first AI universe sim is fast and accurate—and its creators don't know how it works (2019, June 26) retrieved 21 July 2019 from https://phys.org/news/2019-06-ai-universe-sim-fast-accurateand.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.
20 shares

Feedback to editors

User comments

Jun 26, 2019
The first AI universe sim is fast and accurate—and its creators don't know how it works

Why should they? They don't understand how the Universe works, how would they understand a simulation?

Jun 26, 2019
Ya just can't keep yer mouth from driving you over a cliff at more than 85, can ya....

Jun 26, 2019
I imagine the Gods looking upon our universe, scratching their heads and wondering -- How the fuck does that simulation work?

Jun 26, 2019
"its creators don't know how it works"

-I think we are going to be hearing this phrase more and more-

Jun 26, 2019
well. isn't this the point of trying to develop Artificial Stupids?

to attempt to resolve all those complexities beyond the capability of Natural Stupids?

& you all know my low opinion pf Stupid Deities.

Jun 27, 2019
I know how it works.

Jun 27, 2019
Whateth a mistaketh to maketh

Shirley Ho
We used artificial intelligence techniques to simulate the universe
The results are so fast, accurate and robust we aren't sure how it all works

Now there is a turn-up for the books

Shirley Ho, without insulting our intelligence
And that, Shirley Ho
Includes all of us
Because, Shirley Ho
As we are all hominids
We all, presumably
Have the same intelligence
Have a unique technique in insulting what little intelligence we have left

p.s. for Shirley Ho, however intelligent artificial intelligence is, it is not a patch on hominid intelligence

Jun 27, 2019

cant, just cant read in addition to being an anti-science propagandist.

oh deer, poor old granny
had to whip out his/her/it's bigotry & misogyny
waving the hate & fear.
envy & spite,
all over our screens.

the discovery of penicillin was an accident.
however Alexander Fleming was smart enough to recognize that there was something going on that he did not understand.
& made the effort to investigate.

Shirley Ho, as member & leader this group of colleagues.
all of whom participated in this experiment to collaboratively develop a simulation of Universal events that have been observed.

the group has achieved interesting results.
& now will continue the work to figure out what it all means
& what can be accomplished with their results.

granny cant be mature enough to congratulate them.

instead, this pair of losers
have infantile tantrums
of looneytick cult buffoonery.

Jun 27, 2019
Why should they? They don't understand how the Universe works, how would they understand a simulation?


Why do you assume that literally nobody except people on the Thunderbolts forum understand anything about the universe? It's impossible to take you seriously.

Furthermore, they understand the simulations, which is not only implied, but explicitly laid out in the article. What they don't understand is the AI's "remarkable ability to handle parameter variations not found in its training data."

Jun 27, 2019
Do you want Skynet? Because this is how you get Skynet.

Jun 27, 2019
Of the greatest salience, with respect to fields and forces, to this particular board:

"The researchers opted to focus on gravity alone..."

No electricity, no plasma, no magnetism, and... it just   w  o  r  k  s  .

Jun 27, 2019
For Hominidian rrwillsj a complement

rrwillsj
Shirley Ho, as member & leader this group of colleagues.
all of whom participated in this experiment to collaboratively develop a simulation of Universal events that have been observed.

the group has achieved interesting results.
& now will continue the work to figure out what it all means
& what can be accomplished with their results.

granny cant be mature enough to congratulate them

Can you not see, rrwillsj
A complement
When it is staring you in the face, rrwillsj
For if you paused in your red mist
You
Would have read
"for Shirley Ho, however intelligent artificial intelligence is, it is not a patch on hominid intelligence"
For, rrwillsj
We are hominid
For we are more intelligent than artificial intelligence

So you see rrwillsj
A Compliment

Jun 27, 2019
So are the trolls disputing *underlying the science* rather than the new science? More waste of time than usual then:

"The researchers opted to focus on gravity alone because it is by far the most important force when it comes to the large-scale evolution of the cosmos."

And if it was not, if general relativity (gravity theory) did not work, we would have no physical laws - and we would not exist. But maybe trolls don't care for laws or our existence... (LOL, of course they don't, they just want to troll inanely.)

Jun 27, 2019
Everything there is to know is not yet known

Torbjorn_b_g_larsson
And if it was not, if general relativity (gravity theory) did not work, we would have no physical laws - and we would not exist. But maybe trolls don't care for laws or our existence... (LOL, of course they don't, they just want to troll inanely.)

Why
When we do not yet know
That
If what we do know
Happens to be different
Do we have not to exist?

Jun 27, 2019
that penny-ante E/M events occur is not disputed.
but those events need to be considered in context.

that E/M & it's plasma twigs are trite side-effects of Gravitational dominance.

that E/M & plasma are localized, trivial occurrences.
temporarily pretty sparkles
to which the small minded attach way too much unearned importance

yes, you looneyroons simply do not matter at all.
your repetitious dogma & cant are irrelevant to Science, Humanity & the Universe.

Jun 27, 2019
Everything there is to know is not yet known

Torbjorn_b_g_larsson
And if it was not, if general relativity (gravity theory) did not work, we would have no physical laws - and we would not exist. But maybe trolls don't care for laws or our existence... (LOL, of course they don't, they just want to troll inanely.)

Why
When we do not yet know
That
If what we do know
Happens to be different
Do we have not to exist?
says granville

Obviously, torbjorn continues to keep his blinders on so that he fails to see the theoretical/conceptual brilliance of those of us who represent the "new science" that could indeed be the replacement of the "old" or, at the very least, complement that which needs to be complemented.
Just HOW torbjorn comes by the 'knowledge' that humans would not exist, but for GR is a mystery in itself. DO explain your theory, torbjorn.
Of course he will not see my request, as his blinders cause him much ignorance, unless someone else reveals it

Jun 28, 2019
Brain-like system simulates universe effortlessly
Checkmate ametysts

Jun 28, 2019
I dont know if there is any value for mankind mapping out the universe via a machine when humans are unable to validate the results of the experiment by not being able to travel in the universe like star trek. I mean, what the AI is doing is simply creating images of the images we are getting via telescopes. So the AI modeling may in fact be flawed. In any case, my opinion is that this experiment is not about the "accuracy" of the model but how "fast" the computer was able to do it

Jun 28, 2019
... he fails to see the theoretical/conceptual brilliance of those of us...


Let me stop you right there. You've demonstrated your "brilliance" on several occasions, and I don't blame him for rejecting whatever it is you're espousing.

Jun 28, 2019
Everything there is to know is not yet known

Torbjorn_b_g_larsson


Obviously, torbjorn continues to keep his blinders on so that he fails to see the theoretical/conceptual brilliance of those of us who represent the "new science" that could indeed be the replacement of the "old" or, at the very least, complement that which needs to be complemented.

Sorry, SEU, but you do NOT represent new science. New speculation perhaps...
When you've spent the years doing the research that many have, THEN maybe you can lay claim to any science at all...

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more