From Google to galaxy clus­ters

AI has nu­mer­ous uses, from man­ag­ing so­cial me­dia to de­tect­ing warps in space-time

Sky at Night Magazine - - ARTIFICIAL INTELLIGEN­CE -

AI al­go­rithms used by Face­book and Google have been em­ployed by astronomer­s to study a phe­nom­e­non that Al­bert Ein­stein pro­posed in his the­ory of gen­eral rel­a­tiv­ity over 100 years ago. In Oc­to­ber 2017, a team of astronomer­s from the uni­ver­si­ties of Gronin­gen, Naples and Bonn de­vel­oped a method of de­tect­ing grav­i­ta­tional lenses us­ing the same AI as the so­cial net­work and search engine gi­ants.

Grav­i­ta­tional lens­ing is an ef­fect caused by an enor­mous mass warp­ing space-time: mas­sive ob­jects such as galaxy clus­ters can be used as a sort of cosmic mag­ni­fy­ing glass to ob­serve more dis­tant ob­jects.

The AI al­go­rithm the astronomer­s used is called a con­vo­lu­tional neu­ral net­work and works on the same for­mu­lae em­ployed by Google in 2017 to win a game of Go against the world’s best hu­man player. Face­book uses the same al­go­rithms to com­pile data on the im­ages that ap­pear in its users’ time­lines.

The astronomer­s trained their AI us­ing mil­lions of im­ages of grav­i­ta­tional lenses. Nor­mally, hu­man astronomer­s would ex­am­ine all the im­ages to look for po­ten­tial can­di­dates, but the AI was able to find 761 ex­am­ples of grav­i­ta­tional lens­ing in­de­pen­dently in a patch of sky 22 square de­grees across – just over 0.5 per cent of the to­tal area. The astronomer­s then nar­rowed down to 56 can­di­dates await­ing con­fir­ma­tion through tele­scopic ob­ser­va­tion.

Some of the fab­ri­cated im­ages of grav­i­ta­tional lenses that the Gronin­gen, Naples and Bonn astronomer­s used to train their neu­ral net­work to recog­nise the phe­nom­e­non Three of the grav­i­ta­tional lenses dis­cov­ered by the con­vo­lu­tional neu­ral net­work

Newspapers in English

Newspapers from UK

© PressReader. All rights reserved.