The algorithm accelerates the simulation of large and complex universes
Simeon Bird, assistant professor of physics and astronomy at UC Riverside, is a member of a team of astrophysicists who have used machine learning to simulate the universe with high resolution in a thousandth of the time that the methods conventional would take.
The researchers downloaded models of a small region of space at low and high resolution into a machine learning algorithm that is trained to scale the low-resolution models to match the detail of the high-resolution versions. Such learning allows code, which uses “neural networks”, to generate super-resolution simulations containing up to 512 times more particles than low-resolution models.
The work, published in the Proceedings of the National Academy of Sciences, was directed by Yin Li at the Simons Center in New York and Yueying Ni at Carnegie Mellon University. The research paper is titled “AI-Assisted Super-Resolution Cosmological Simulations”.
Bird, who joined UCR in 2018, studies machine learning, black holes, neutrinos, and dark matter. He said it was a privilege to collaborate on the project. It retained the simulation code used to generate the training data. In this Q&A, he answers a few questions about the project:
Q. What is an artificial intelligence neural network and how does it work?
A neural network is a very flexible model to adapt to any type of data. You can think of it as a series of filters that show different interesting characteristics of the input. Neural networks are trained to select specific interesting parts of the simulation and reproduce them. For our work, this is done by forming a network that tries to reproduce the simulation and a network that tries to find differences between the reproduction and the original. By playing these two networks against each other, we end up with something very difficult to distinguish from the original simulation.
Q. How did you get involved in this project?
I run a lot of very large computer simulations. This takes lots of time. Nowadays it is very difficult to make simulation much faster. This type of machine learning provides the ability to resize simulations to ten times their current size, which would be very difficult any other way.
Q. What do you think this technology will make possible for astronomical research?
This technology will eventually allow much larger simulations. These simulations are needed in the near future to ensure that we have models to compare to future much larger astronomical surveys.
Q. How does the algorithm learn to scale low resolution models to match details found in high resolution versions?
It is quite difficult to explain this simply. We think it works because on a large scale one part of the cosmic web looks a lot like another: once dark matter begins to crumble, it forgets where it came from.
Q. The code can take large-scale low-resolution models and generate super-resolution simulations containing up to 512 times the number of particles. How can you be sure that moving upmarket does not generate unobservable “nonsense”?
For this model, we were able to run a simulation directly with 512 times more particles and verify that it looks like the output of upscaling. When we start using this on bigger issues, where we can’t do that – that’s the point! If we can just run the larger simulation, there is no need for machine learning – we’ll run smaller simulations of parts of the simulation on a large scale and verify them.
Q. The team could not run the simulation generator for two years. What were the obstacles?
Machine learning is a rapidly evolving field. Over the past two years, progress has been made in how to train these models. We applied these advances and the problem of previously impossible training suddenly became possible.
Q. Where else can this new technology potentially be used?
This new technology can, hopefully, dramatically increase the dynamic range of our simulations, allowing us to model individual galaxies along with the large-scale distributions of galaxies in the sky.
Q. How can someone use this technology?
This technology is available free of charge, based on open source technology. Anyone with a decently powerful computer and a graphics card could do something similar with enough patience.
Miniature photo: Milky Way photographed by the Spitzer telescope. (NASA / JPL-Caltech / S. Stolovy; Spitzer Science Center / Caltech)