Becoming all distorted images and funny names, it can be easy to forget that machine learning can have important applications in science – especially when it comes to things categorize. Scientists have recently used a neural network to identify distant galaxies.
An international research team points out that there are a variety of space images, both from the near and the distant universe. But around the corner are more polls that will have tons more data – more than people can effectively scour. It can be difficult to synthesize this data and connect the dots between young and old galaxies. This is where neural networks come into play.
"Once we've trained a computer with thousands of images from our simulations, computers can see things we just can not see," says Joel Primack, Professor of Physics at the University of Washington, California. Santa Cruz, said Gizmodo. "That's very helpful."
The researchers began with a powerful simulation to create 35 model galaxies and then used other software to create around 10,000 images that are both clear and blurred. They trained a neural network in the pictures to identify their similarities. The researchers then fed the trained network with real data images of distant galaxies from the CANDELS survey. It has successfully divided the galaxies into three categories based on their shape. These categories correspond to three phases of galactic evolution, which they call the phase before the blue nugget, the blue nugget phase, and the blue nugget phase.
After feeding the images of the neural network of simulated galaxies, the researchers were able to obtain useful information about real galaxies. That's pretty cool.
A neural network would obviously be very helpful for large scale surveys. The Wide Field Infrared Survey Telescope, which would launch in the 2020s, could capture millions of galaxies with Hubble's resolution in frames . Every night from Earth, the Large Synoptic Survey Telescope will map a large part of the southern sky and record 15 terabytes of data daily. A neural network could quickly identify things that are of the utmost interest to astronomers, or point out things that a human eye might miss.
Others are excited. "There is a hope among some researchers that when artificial intelligence can sort, classify, and report on interesting things, the human capacity to learn about the universe is extended beyond what we can imagine by itself "said Michael Oman-Reagan, Ph.D. student at Memorial University in Newfoundland and Labrador, Canada, exploring beyond the solar system and the potential for extraterrestrial life.
Maybe machine learning could help people find extraterrestrial intelligence, he thought.
That's exciting stuff, but Primack warned me to be careful – that's just proof of the concept. These training sets could take a long time to generate. Therefore, they are not a realistic way to categorize the data. In addition, the simulations may still be too limited to fully capture the diversity of galaxies, according to the paper published in the Astrophysical Journal.
But things are progressing, and Primack is not the only team working on it. Others also train computers in large galactic datasets, and Primack has given a reputation for collaborating on feedback in Realistic Environments that create realistic galactic models.
Ultimately, it should not be the job of the AI to replace scientists, but to help them with the incredible amounts of data recorded by the newest observatories.