Scientists use AI to accelerate universe modeling by 1,000 times

Researchers at Carnegie Mellon University have developed artificial intelligence that models the universe on a single graphics processing unit 1,000 times faster than existing methods.

To create the algorithm, the researchers combined modern approaches to generation and quality improvement of images using neural networks. Their use is said to save time, energy and computing power, according to the experts.

The scientists noted that for simulating a region of the Universe spanning about 500 million years, containing 134 million particles, existing methods require 560 hours. The algorithm they developed handles the task in 36 minutes.

The researchers repeated the experiment, expanding the area by 1,000 times to 134 billion particles, and found that the method they created generates an image in 16 hours on a single GPU.

“Using existing methods, completing the modeling of such size and resolution would take months for a specialized supercomputer,” the developers added.

They say this does not mean that AI “knows” what the Universe looks like beyond our reach. The new method enables upgrading low-resolution simulation images to high resolution and allows scientists to generate them using less time, energy and computing resources.

In April 2021, developed a model of explainable artificial intelligence, which can shed light on how machine learning makes decisions.

In March, Nvidia created a machine-learning algorithm, accelerating human genome analysis by 100x.

Subscribe to ForkLog news on Telegram: ForkLog AI — all the latest in AI.