Can Artificial Intelligence Accelerate Astrophysics?
New AI machine learning generates cosmic simulations with super-resolution.
Posted May 6, 2021 | Reviewed by Jessica Schrader
A new study in the Proceedings of the National Academy of Sciences of the United States of America (PNAS) shows how artificial intelligence (AI) deep learning can achieve cosmological simulations with super-resolution to accelerate a greater understanding of our universe.
The study’s lead author is Yin Li, a joint research fellow in astrophysics and computational mathematics at the Simons Foundation’s Flatiron Institute. The co-authors include Carnegie Mellon University researchers Yueying Ni, Rupert Croft and Tiziana Di Matteo, Simeon Bird at the University of California, Riverside, and Yu Feng at the University of California, Berkeley.
“Cosmological simulations are indispensable for understanding our universe, from the creation of the cosmic web to the formation of galaxies and their central black holes,” wrote the scientists. “This vast dynamic range incurs large computational costs, demanding sacrifice of either resolution or size and often both.”
In astrophysics, cosmological simulations are essential for aiding scientists in formulating testable predictions to prove or disprove by telescopic observations.
“Even with supercomputers we are forced to decide whether to maximize either resolution, or volume, or else compromise on both,” the scientists wrote.
The researcher team used AI deep learning, a subset of machine learning, to transform low-resolution images. For inputs and outputs, the deep neural network model used the displacements of particles in the N-body simulations. In astrophysics, N-body simulations are a commonly used tool to show the dynamical system of particles.
Using a Generative Adversarial Network (GAN)-based algorithm, the researchers produced a variety of super-resolution versions of different low-resolution images.
“We build a deep neural network to enhance low-resolution dark matter simulations, generating super-resolution realizations that agree remarkably well with authentic high-resolution counterparts on their statistical properties, and are orders-of-magnitude faster,” the researchers wrote. “It readily applies to larger volumes, and generalizes to rare objects not present in the training data.”
According to the researchers, they were able to “enhance the simulation resolution by generating 512 times more particles,” as well as predict displacements from the initial positions. Additionally, the study shows that super-resolution simulations that are more than a thousand times larger than the training sets can be generated using the AI method.
“As telescopes and satellites become more powerful, observational data on galaxies, quasars and the matter in intergalactic space becomes more detailed and covers a greater range of epochs and environments in the universe,” the researchers wrote.
The scientists envision using artificial intelligence may help enable the prediction of the properties of supermassive black holes, quasars, and galaxies that is much faster computationally, yet statistically on par with full-scale hydrodynamic models.
Artificial intelligence machine learning is rapidly being deployed as a tool to speed up scientific research. Now with this demonstrated proof-of-concept, astronomers and physicists have a new method to model and simulate our universe across an unprecedented dynamic range in the future.
Copyright © 2021 Cami Rosso. All rights reserved.