Una collaboración internacional lederada por astrofísicos de la Autonomous University of Madrid (UAM) has estimated, for the first time, by means of artificial intelligence techniques, the total material content of a large set of galaxy clusters. Los resultados, publicados en la revista Nature Astronomy.
The automatic learning algorithm used is based on convolutional neural networks, a revolutionary model developed for automatic image processing. “The procedure consists of compressing the information contained in the images, so that it can be easily compared with some internal property of the image, which in this case would be the total mass of the cloud that occupies the image,” he explains Daniel de Andrésprofesor ayudante del departmenta de Física Teórica de la UAM y primer firmante del trabajo.
Para trainar e-mail algorithmlos investigadores tuvenor que generar casi 200,000 images equivalent to how the Planck satellite would observe a set of numerically simulated cumulus clouds. These synthetic images were obtained from the results of the numerical simulations of the project The Three Hundred, led also by the researchers of the UAM.
“Una vez que la network convolutional fue entrenada, se le aplicó a más de mil imágenes reales de cúmulos de galaxias observed by e-mail Satellite Planckgiving as a result the estimation of the total mass of the object, according to what the algorithm has learned from the simulations, but in this case, free of possible biases due to the theoretical hypotheses that are assumed in previous studies”, explains Daniel de Andrés.
Por su parte, el coautor Weiguang Cui, affirms that “los resultados obtenidos son muy excitantes. The artificial intelligence is a tool that will help us to understand the complex relationships between the different components of matter that occupy the universe”. In collaboration, researchers from the University of Edinburgh, the MIT (EE UU) and the University of Sapienza (Rome), together with experts from EURA NOVA (Belgium), a company specialized in automatic learning techniques.
Aprendizaje automático y astrofísica
La intelligentia artificial, y en particular e automatic learning. In this form, los ordenadores pueden realizar tasks complejas, como recognar objetos en imágenes o jugar al jedrez.
El aprendizaje automático has given place recently to multiple applications in different fields of industry and research. Not only is it useful to speed up the analysis of huge amounts of data, it also brings new discoveries. In particular, the methods of artificial intelligence for the analysis of astrophysical data are being used every time in the most habitual way in the processing of astronomical images.
Given that astrophysical objects can only be detected from electromagnetic radiation emitted at different frequencies by interstellar gas and stars (from gamma rays to radio waves), the information that can be derived corresponds only to the matter that makes up the objects that emit light. Pero resulta que, cuanto major es un objeto astrónómico, su contenido total de materia está más dominated por ‘materia oscura’, que no emite luz. This is the case of the most massive objects in the universe, the clusters of galaxies.
Cúmulos de galaxias y materia oscura
Los cúmulos de galaxias son agrupaciones de centenares o miles de galaxias unidas gravitationally. The total matter that generates this gravitational field is formed, in more than 85%, by a component of nature that we still do not understand, and that is called materia oscura.
Hasta ahora, la materia total de estos objetos se había medido principalum de dos maneras. A partir de observaciones de la emission en x-rays due to intracumular gas. Y mediate las distortiones del frequency spectrum de los photones de la radiación de fundo cósmica, cuando atraviesan el gas caliente del cúmulo en su recorrido hacia nuestros radiotelescopios.
“In both cases, from the two-dimensional images that we receive with our instruments, we have to apply certain theoretical hypotheses on the state of the cumulus gas to derive the total mass that it contains,” explains Daniel de Andrés. “The only form that we had until now to establish these theoretical relations between the emission of gas and the mass was based on numerical simulations that tried to reproduce the formation of these objects and that modeled the complex relations that exist between the various components of the cluster: gas, estrellas, supernovas, agujeros negros y materia oscura”, concluded the researcher.