A computer program has learned to distinguish the galaxy like a man


10/06/2010

Computer algorithm, which served as a model for the human brain has learned to recognize different types of galaxies, from the spiral and elliptical ending. Now, he is able to help their colleagues - stargazers of flesh and blood, scrupulously fulfill exhausting job classification billion galaxies.

This self-learning program that is able to correctly identify the type of the galaxy in 90 percent of cases, according to scientists from Univsiteta London (University College London) and Cambridge (University of Cambridge) in the UK.

The program should help scientists cope with the endless stream of images of galaxies, which come from projects such as the Sloan Digital Sky Survey and the Galaxy Zoo. Billions of galaxies in the known universe, we can take many forms, they are elliptical, spiral, irregular shape.

"Telescopes are the next generation, which is now under construction, will create the image of hundreds of millions and even billions of galaxies in the next ten years" - said Manda Banerji (Manda Banerji), an astronomer at the University of Cambridge. "The numbers are discouraging, it is physically impossible to check them all with the human eye."

The classification of types of galaxies - is the first step toward understanding the origin and evolution of galaxies.

More than 250,000 people have helped astronomers to classify 60 million galaxies in the online project Galaxy Zoo. Astronomers have used the classification of Galaxy Zoo, learning a computer algorithm that is iskucstvennoy neural network.

Iskucstvennaya neural network is able to analyze the complex relationships between different variables, such as shape, size, color of astrophysical objects, and to do on the basis of this conclusion about the type of the analyzed galaxies. This algorithm is similar to the biological neural networks of living beings.

Before you test the capabilities of the program, astronomers have spent training a computer algorithm, using the 75,000 already classified astrophysical objects taken from the project Galaxy Zoo. The algorithm then classifies 1,000,000 objects. On the basis of the tests, the researchers made the setting of some parameters of the program, which helped to achieve the level of correct identification of type galaxies in 90% of cases.

The creators of the program, also met with some difficulties. For example, the algorithm is difficult to give the definition of types of red spirals and blue ellipticals. This problem is due to an insufficient number of examples of such objects - the program simply does not have enough training material.

The study also does not include intermediate types of galaxies, for example consisting of several merged galaxies, which are not easy to categorize.

But even in this imperfect form, the automatic classification of the types of galaxies, astronomers will, buried in the flow of visual information about galaxies, a great help. In a recent study of the sky, it was found that astronomers have underestimated the number of distant galaxies by 90%.

The human eye - are very effective in recognizing patterns. Intelligent computing techniques that allow us to reproduce this behavior, it is very important as we begin to expand the boundaries of the visible universe and discover more distant galaxies, "- said Ofer Lahav (Ofer Lahav), an astrophysicist at the University of London." This study - an important step in that direction. "All the details about the main events of the day widgets news let us know.

Original: Space.com


Geminids meteor rain held this week
Satellite-zombies come back to life
NASA will launch nanosatellites running smartphones
In the universe is three times more stars than previously thought
Astronomers from the United States opened a 25 "mini-supernova"