AI smashes barriers to computational resources and time.

A new AI application makes it dramatically easier to simulate a vast, complex universe in high resolution. This breakthrough is a huge advance in high-resolution cosmological simulations, reducing computation time from hundreds of hours to just a few minutes. In a study published online May 4 in Proceedings of the National Academy of Sciences, researchers at the Flatiron Institute in New York City said that conventional methods were unable to simulate the universe at both high resolution and large volume. (Science Daily)

“With our new technique, it’s possible to have both efficiently,” said study lead author Yin Li, an astrophysicist at the Institute. “In the future, these AI-based methods will become the norm for certain applications.”

Flatiron Institute’s method: How it works

Li and his colleagues trained a machine-learning algorithm to simulate a low-resolution model of a small region of space to a higher resolution by feeding it both kinds of models.

The algorithm learned how to upscale the low-resolution model and generate the kind of detail usually found in high-resolution versions.

The researchers used neural networks that could predict how gravity would move dark matter around over time after being supplied with training data. The networks ran calculations using the information and then simulated the high-resolution models. These results were then compared with the expected high-resolution models. With more training, the networks became progressively more accurate.

The process pioneered at the Flatiron Institute generated dramatic savings in time. For example, existing methods would need 560 hours to create a high-resolution simulation of a region in the universe approximately 500 million light-years across and containing 134 million particles. With the machine learning algorithm, the researchers completed the exercise in just 36 minutes.

“We couldn’t get it to work for two years,” Li says, “and suddenly it started working. We got beautiful results that matched what we expected. We even did some blind tests ourselves, and most of us couldn’t tell which one was ‘real’ and which one was ‘fake.’”

Related Story:      AI to Help Analyze Large Images from Aerial Sources