Exoplanet detection with Genesis

Convolutional Neural Networks (CNNs) have shown to o®er a consistent and reliable foundation for theautomatic detection of potential exoplanets. CNNs rely on an abundance of parameters (over-parameterization) to achieve their impressive detection performances. Astronet was one of the ̄rst CNNsfor exoplanet detection. It takes as input folded lightcurves in two views: a local view (the transit) and aglobal view (the entire orbital period including the transit). A more recent CNN called Exonet-XS im-proved on Astronet's performance while having considerably less parameters, thereby reducing the risk ofover ̄tting. Exonet-XS also uses two views as input. In this paper, we propose Genesis, an even moresimpli ̄ed CNN for exoplanet detection from folded lightcurves using only one view. In addition, we proposeto use a more reliable validation procedure that is custom in CNN-based exoplanet detection studies: theMonte Carlo Cross-Validation (MCCV) procedure. We show that the use of MCCV improves the reliabilityof the estimation of the detection performance by providing a (discretized) probability distribution, ratherthan a point estimate. Using MCCV we show that Astronet with only one view performs on a par with theoriginal two-view version. More importantly, our fair comparative evaluation (without stellar parametersand centroids) reveals that Genesis outperforms Exonet-XS and Astronet. We conclude by stating thatexisting exoplanet detection CNNs are too complex for the task at hand and that future evaluations ofperformances should use MCCV or similar validation procedures