A new study recently published in The Astrophysical Journal explores the possibility of inferring the properties of the Galactic neutron star population through machine learning – an implementation of artificial intelligence that gives systems the ability to automatically improve and learn from previous experiences without being explicitly told how to do so.

ICE researchers Michele Ronchi, Vanessa Graber, Nanda Rea and Alberto García, who are members of the ERC Magnesia project, explored for the first time how machine learning can be used for the comparison between the synthetic sample and the observed characteristics. The article led by ICE researchers found out that a convolutional neural network is able to estimate with high accuracy the parameters which control the current positions of a mock population of pulsars.

Although about a billion neutron stars are expected to exist in our own galaxy, observational constraints limit us to only detecting a small fraction of them. That is why we only know a few thousands of these compact objects to date. To overcome this gap, so-called population synthesis approaches are used to theoretically model the full population of pulsars in the Milky Way.
 
The analysis further highlights the crucial need for increasing the sample of known pulsars and accurately classifying them in order to infer the properties from the real pulsar population. This is closely linked to the science drivers of upcoming radio telescopes, in particular the Square Kilometer Array, an international effort to build the world’s largest radio telescope.

Further implications

Machine learning techniques have seen a lot of interest in the astronomy and astrophysics community, where it is often no longer possible to evaluate large amounts of data by hand.

This kind of global population synthesis study using machine learning allows scientists to better constrain the input physics and learn more about neutron stars on an individual level.