As you know, they used math to discover at least one planet and several other astronomical objects instead of relying on on observations to expand their knowledge. Recently, researchers used a machine learning algorithm to look one simulated galaxy and from that they were able to predict the makeup of the whole digital universe the galaxy is in. This is like looking at a grain of sand and using that information to calculate the mass of a moon.
This discovery will someday allow cosmologists to make conclusions about the cosmos based on it's basic building blocks. What this does is it makes it easier to study the universe because you only have to examine one galaxy to extract enough information to make accurate conclusions about the whole universe.
This developed from a challenge. The challenge was to create a neural network based on the properties of a galaxy that could estimate a couple of cosmological attributes. This challenge was given so the person could become more familiar with machine learning. In the process, it was noticed that the program properly calculated the density of matter.
The neural network itself analyzed one million simulated galaxies for size, composition, mass, and other characteristics, but this research only used 2,000 digital universes created by the program. The neural network was able to connect the information to the matter in the parent universe. Most of these universes were composed of between 10 percent and 50 percent visual matter with the rest being dark matter. In the simulations, the visible matter and dark matters swirled together forming galaxies. The simulation even applied events such as supernovas or jets erupting from black holes.
Furthermore, the neural network was able to predict cosmic density within 10 percent regardless of the galaxy and this result was quite unexpected since galaxy by their nature are quite chaotic. No matter what type of galaxy, they were able to keep up with the overall density of matter. This leads to the idea that both universes and galaxies are simpler than previously thought.
This then allowed researchers to analyze the neural network to make sure it wasn't getting the data from coding of the simulation and discovered the neural network used things like the property associated with the rotational speed that determines the type of matter found in the central zone. In addition, the neural network is able to look at multiple properties at once rather than one or two at a time like humans.
Scientists are not sure which properties the neural networks are able to connect with which galaxies or universes yet since this is just in the beginning stages but the process does open up some interesting lines of investigation. Let me know what you think, I'd love to hear. Have a great day.