New Technique Models the Universe in a Fraction of the Time

Researchers use machine learning to transform a low-res model for a high-res simulation.

A new technique developed at the Flatiron Institute will allow astrophysicists to model the universe in a fraction of the time it used to take. According to lead author Yin Li, “At the moment, constraints on computation time usually mean we cannot simulate the universe at both high resolution and large volume. With our new technique, it is possible to have both efficiently.”

Li and his collaborators used machine-learning techniques to transform a low-resolution model into something closer to the precision of a high-resolution model. They fed the machine-learning algorithm a low resolution model for a small sector of space and then gave it a higher resolution model for the same sector. The system was then able to train itself to build a high-resolution model based on low-resolution input. After that, the researchers were able to input a low-res models for much larger sectors, and the algorithm was able to generate “super resolution” simulations with over 512 times as many particles.

The announcements adds, “For a universe 1,000 times as large with 134 billion particles, the researchers’ new method took 16 hours on a single graphics processing unit. [According to Li]...existing methods would take so long that they wouldn’t even be worth running without dedicated supercomputing resources.