Because we cannot travel back in time, one of the most comprehensive methods to understand how the universe developed is to design computer simulations of the process using the already known things about it.
An international team of scientists has now created an AI that can rapidly produce highly-accurate, three-dimensional simulations of the universe.
The team of researchers explains how they designed the universe simulator, which they have called the ‘Deep Density Displacement Model (D3M), in a study issued in the journal Proceedings of the National Academy of Sciences. The aim was to teach the D3M how to pattern the way gravity models the universe. To do so, the team started implementing in the system 8,000 different simulations focused on gravity designed by a highly accurate existing universe simulator.
That system required 300 computation hours to design just one of its simulations, but after having access to the data, D3M was capable of generating its own simulations of a cube universe 600 million light-years overall in only 30 minutes. The simulations were more precise than those of the already existing ‘fast’ systems, which require a few minutes to design a simulation.
Speed is not the only outstanding feature of D3M. Another thing that makes it remarkable is the ability to precisely simulate what the universe would look like even if the scientists changed parameters that were not incorporated in its training data. For instance, they would adjust the percent of dark matter in their universe, and D3M could precisely simulate that universe’s development.
Besides helping researchers understand the universe’s development, this peculiar behavior has the capacity to aid computer scientists to gain a better understanding of AI.
Researcher Shirley Ho said in a press release that they can be an unusual practice place for the machine learner to use to see why the model computes so well. This is a two-way street forming between science and deep learning.