Bonn 2020 – scientific programme

The DPG Spring Meeting in Bonn had to be cancelled! Read more ...

Parts | Days | Selection | Search | Updates | Downloads | Help

EP: Fachverband Extraterrestrische Physik

EP 13: Exoplanets and Astrobiology

EP 13.4: Talk

Friday, April 3, 2020, 11:00–11:15, H-HS VIII

Classifying Exoplanet Candidates with Convolutional Neural Networks: Application to the Next Generation Transit Survey — •Alexander Chaushev¹ and Liam Raynard² — ¹TU Berlin, Berlin, Germany — ²University of Leicester, Leicester, UK

A key bottleneck in the discovery of transiting exoplanets is the large number of false positives produced by existing detection algorithms. Currently the solution to this problem is to vet the candidates by hand, however this is time consuming and can be inconsistent. Recently convolutional neural networks (CNNs), a type of `Deep Learning' algorithm, have been shown to be effective at this task [Shallue+19].

Here I will present results from the on-going effort to automate the Next Generation Transit Survey (NGTS) candidate vetting process using a CNN. Currently we are able to exclude ~50% of false positives, while recovering ~90% of our manually identified candidates and all currently known planets in the NGTS dataset [Chaushev+19]. A key goal of the project is to understand and improve the network performance in the lowest signal to noise regimes. In this regard, NGTS provides a unique dataset as it has been continually pushing to find planets on the edge of detectability, leading to the discovery of NGTS-4b the shallowest transit discovered from the ground to date [West+19]. This makes NGTS an ideal testing ground for CNNs, and improvements made in the techniques here can readily be applied to space based data from K2 and TESS currently, and PLATO in the future.