https://www.eurekalert.org/multimedia/pub/web/88327_web.jpg

#

Through his observations man has learned a lot about planets and their origins and has also developed procedures that can predict such a genesis. With the help of AI, Swiss scientists have now been able to calculate the origin of planets much faster and more accurately.

The birthplace of planets are ring-shaped accumulations of dust and gas around young stars, also known as protoplanetary discs. It is not yet known in detail how the first planet germs develop from this dust, whether the initial nuclei form rock planets like the earth or gas planets like Jupiter and what it takes to form different planets. It is assumed, however, that it naturally depends on the material in the dust rings, the pressure, the temperatures and the conditions of the star that irradiates the circumstellar disk.

Until now, astronomers had to solve a series of complex differential equations to simulate the growth process and inner structure of planets, which was a very time-consuming procedure. Researchers have now presented a new process that greatly abbreviated these calculations with the help of AI.

#

https://www.researchgate.net/profile/Christian_Rab/publication/321488036/figure/download/fig1/AS:567688634535936@1512358925723/Sketch-of-a-typical-protoplanetary-disk-and-the-various-line-emitting-regions-in.png

#

The researchers from the Bern University and from the International Space Science Institute in Bern used the AI variant DeepLearning for this purpose. Today, this variant of artificial intelligence is used for image recognition or automatic speech translation. The researchers simply had to create a database with millions of different possibilities of planetary structures and then feed it to a neural network. The researchers simply put in a set of algorithms that reads the mathematical calculations and has the ability to learn without being programmed, for this purpose.

The neural network learned how to quickly shorten the time-consuming calculations thanks to a short training based on the database. The network is now able to predict the mass of a planet with certain conditions with very good accuracy. This is even much faster if you have to solve a differential equation.

This new method with the use of DeepLearning is much more precise than other developed methods that used analytical formulas to solve differential equations. However, these analytical formulas had imprecise values and predicted, for example, that a planet should grow to the mass of one Jupiter, but in reality it cannot even have the mass to do so. This misunderstanding is now fixed and requires much less time and effort.

#

https://supernovacondensate.files.wordpress.com/2012/07/soot-line1.jpg?w=1024

#


https://steemitimages.com/0x0/https://steemitimages.com/0x0/https://cdn.steemitimages.com/0x0/https://steemitimages.com/DQmaejvzeBGuF9R2XvtBzuxwnismQna9u2kC3anmgvkzhFf/trennlinie_umgekehrt.png<center><sub>source of Informations: 
[arXiv.org](https://arxiv.org/abs/1903.00320) - [researchGate](https://www.researchgate.net/publication/331485375_Using_Deep_Neural_Networks_to_compute_the_mass_of_forming_planets)
Images: [1](https://www.eurekalert.org/multimedia/pub/web/88327_web.jpg), [2](https://steemitimages.com/0x0/https://www.researchgate.net/profile/Christian_Rab/publication/321488036/figure/download/fig1/AS:567688634535936@1512358925723/Sketch-of-a-typical-protoplanetary-disk-and-the-various-line-emitting-regions-in.png), [3](https://supernovacondensate.files.wordpress.com/2012/07/soot-line1.jpg?w=1024)</center></sub>
#
https://steemitimages.com/DQmNhHzhYBaCtqdK4FRJUjnY1ouSv5UDeCfR2tNHSmMqyA9/trennlinie.png
<center>https://steemitimages.com/DQmdi8fsmuDprrMivfVWRbWSDFGSwtHfpS7UR9CK9MD6tXT/new%20-%20logo%20-%20without%20shade.gif</center>

https://steemitimages.com/0x0/https://cdn.steemitimages.com/0x0/https://steemitimages.com/DQmaejvzeBGuF9R2XvtBzuxwnismQna9u2kC3anmgvkzhFf/trennlinie_umgekehrt.png