Discovering pulsars with machine learning

Pulsars are compact and fast-spinning stars whose unique properties allow us to probe the laws of Physics in truly extreme conditions and to put Einstein’s General Relativity to the test. New pulsars are usually discovered with a radio telescope, but the radio band is filled with artificial signals reproducing a pulsar’s signature. Here we present a new classification algorithm that differentiates pulsar signals from the rest as accurately but thousands of times faster than a human expert. We used it to discover 21 new pulsars, half of them left behind in data previously searched with more conventional methods. Such algorithms will be critical to the success of pulsar searches with the upcoming Square Kilometer Array.