Classifying exoplanet candidates with convolutional neural networks: application to the Next Generation Transit Survey

Abstract

Vetting of exoplanet candidates in transit surveys is a manual process, which suffers from a large number of false positives and a lack of consistency. Previous work has shown that Convolutional Neural Networks (CNN) provide an efficient solution to these problems. Here, we apply a CNN to classify planet candidates from the Next Generation Transit Survey (NGTS). For training datasets we compare both real data with injected planetary transits and fully-simulated data, as well as how their different compositions affect network performance. We show that fewer hand labelled lightcurves can be utilised, while still achieving competitive results. With our best model, we achieve an AUC (area under the curve) score of (95.6 ± 0.2)% and an accuracy of (88.5±0.3)% on our unseen test data, as well as (76.5±0.4)% and (74.6±1.1)% in comparison to our existing manual classifications. The neural network recovers 13 out of 14 confirmed planets observed by NGTS, with high probability. We use simulated data to show that the overall network performance is resilient to mislabelling of the training dataset, a problem that might arise due to unidentified, low signal-to-noise transits. Using a CNN, the time required for vetting can be reduced by half, while still recovering the vast majority of manually flagged candidates. In addition, we identify many new candidates with high probabilities which were not flagged by human vetters.

NGTS 15b, 16b, 17b, and 18b: four hot Jupiters from the Next-Generation Transit Survey

“…The initial fits to the NGTS data provided by revealed that the depths, widths and shapes of the transits for each object were compatible with transiting hot Jupiter planets. In addition, a convolutional neural network applied to the NGTS data found that the probabilities of each lightcurve containing a transiting exoplanet were all greater than 0.95, consistent with previous confirmed NGTS planet discoveries (Chaushev et al 2019).…”

NGTS-14Ab: a Neptune-sized transiting planet in the desert

“…We also note that this system received a high planetary probability of 0.97 from a neural network trained to distinguish between transiting planetary systems and false positives in NGTS data (Chaushev et al 2019).…”

Machine-learning Inference of the Interior Structure of Low-mass Exoplanets

“…In this work, we explore a new approach to the interior characterization of exoplanets by employing a deep learning neural network to treat this inverse problem. In recent years, deep neural networks have been used in a number of exoplanetary science studies, with applications ranging from the detections of planetary transits (Pearson et al 2018;Shallue & Vanderburg 2018;Chaushev et al 2019) to atmospheric composition retrieval from measured planet spectra (Zingales & Waldmann 2018;Márquez-Neila et al 2018), and the computation of critical core and envelope masses of forming planets (Alibert & Venturini 2019). We approach the characterization of exoplanets by first creating a large dataset of synthetic planets and then training a multitask (Caruana 1997) mixture density network (MDN, Bishop 1994) to infer plausible thicknesses of the planetary layers using observables like mass and radius.…”