Improved Photometric Classification of Supernovae using Deep Learning

Abstract

We present improved photometric supernovae classification using deep recurrent neural networks. The main improvements over previous work are (i) the introduction of a time gate in the recurrent cell that uses the observational time as an input; (ii) greatly increased data augmentation including time translation, addition of Gaussian noise and early truncation of the lightcurve. For post Supernovae Photometric Classification Challenge (SPCC) data, using a training fraction of $5.2\%$ (1103 supernovae) of a representational dataset, we obtain a type Ia vs. non type Ia classification accuracy of $93.2 \pm 0.1\%$, a Receiver Operating Characteristic curve AUC of $0.980 \pm 0.002$ and a SPCC figure-of-merit of $F_1=0.57 \pm 0.01$. Using a representational dataset of $50\%$ ($10660$ supernovae), we obtain a classification accuracy of $96.6 \pm 0.1\%$, an AUC of $0.995 \pm 0.001$ and $F_1=0.76 \pm 0.01$. We found the non-representational training set of the SPCC resulted in a large degradation in performance due to a lack of faint supernovae, but this can be migrated by the introduction of only a small number ($\sim 100$) of faint training samples. We also outline ways in which this could be achieved using unsupervised domain adaptation.