A team of astrophysicists developed an AI universe simulator that can generate complex 3D models of the universe.
It’s the first time that artificial intelligence technology has been used to perform such a task.
The project, called Deep Density Displacement Model or D3M, is the brainchild of Siyu He, a researcher from Flatiron Institute‘s Center for Computational Astrophysics, together with her colleagues from Carnegie Mellon University, UC Berkeley, Kavli Institute, and the University of British Columbia.
In a paper published in the journal Proceedings of the National Academy of Sciences, the scientists described how fast and accurate D3M that they can’t explain how it works. Shirley Ho, a co-author of the study from Flatiron Institute, said:
“We can run these simulations in a few milliseconds, while other ‘fast’ simulations take a couple of minutes. Not only that, but we’re much more accurate.”
First AI Universe Simulator is Eerily Accurate
Despite the impressive speed of the AI system in producing simulations, it’s not what surprised the astrophysicists.
Instead, He and her team were perplexed with how their AI universe simulator can still create models of the universe accurately when certain perimeters were tweaked, even without receiving any training data about those varied parameters.
Ho added:
“It’s like teaching image recognition software with lots of pictures of cats and dogs, but then it’s able to recognize elephants. Nobody knows how it does this, and it’s a great mystery to be solved.”
For instance, if tweaked to show how much dark matter is present in the cosmos, the AI universe simulator can produce a precise 3D model with that information.
The astrophysicists trained D3M’s deep neural network using 8,000 different simulations produced by one of the highest-accuracy computer models available today.
He and her colleagues’ first simulations using D3M were of a box-shaped universe 600 million light-years across.
By comparison, D3M was able to complete one simulation in just 30 milliseconds while the slow-but-accurate technique took hundreds of computational hours to complete one simulation and current fast simulators took minutes.
Also, D3M’s results only had a relative error of 2.8 percent, whereas the existing high-accuracy model had a 9.3 percent relative error. Ho noted:
“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. It’s a two-way street between science and deep learning.”
Comments (0)
Most Recent