SuperNNova: an open-source framework for Bayesian, neural network-based supernova classification

Abstract

We introduce SuperNNova, an open source supernova photometric classification framework which leverages recent advances in deep neural networks. Our core algorithm is a recurrent neural network (RNN) that is trained to classify light-curves using photometric information only. Additional information such as host-galaxy redshift can be incorporated to improve performance. We evaluate our framework using realistic supernovae simulations that include survey detection. We show that our method, for the type Ia vs. non Ia supernovae classification problem, reaches accuracies greater than 96.92 ± 0.09 without any redshift information and up to 99.55 ± 0.06 when redshift, either photometric or spectroscopic, is available. Further, we show that our method attains unprecedented performance for classification of incomplete light-curves, reaching accuracies > 86.4 ± 0.1 (> 93.5 ± 0.8) without host-galaxy redshift (with redshift information) two days before maximum light. In contrast with previous methods, there is no need for time-consuming feature engineering and we show that our method scales to very large datasets with a modest computing budget. In addition, we investigate often neglected pitfalls of machine learning algorithms. We show that commonly used algorithms suffer from poor calibration and overconfidence on out-of-distribution samples when applied to supernovae data. We devise extensive tests to estimate the robustness of classifiers and cast the learning procedure under a Bayesian light, demonstrating a much better handling of uncertainties. We study the benefits of Bayesian RNNs for SN Ia cosmology. Our code is open-sourced and available in github ‡.

Supernova host galaxies in the dark energy survey: I. Deep coadds, photometry, and stellar masses

“…As transient surveys probe larger areas with deeper observations, however, it not feasible to classify all of the SNe spectroscopically. We thus define samples by classifying SNe 'photometrically', principally using the light-curve shape and colour to distinguish SNe Ia from core-collapse events using classifiers such as pSNid (Sako et al 2008), SuperNNova (Möller & de Boissière 2019), and RAPID (Muthukrishna et al 2019).…”

“…Redshifts from the hosts improve the photometric classification of transients (e.g. Olmstead et al 2014;Sako et al 2014), with classification accuracy of the SuperNNova classifier improving from 97 per cent to >99 per cent with the addition of redshift (Möller & de Boissière 2019). Secondly, even after brightness corrections are applied using known correlations in their light-curve shape (stretch) and colour, a residual intrinsic scatter in their absolute peak brightness is still measured.…”

Background rejection in atmospheric Cherenkov telescopes using recurrent convolutional neural networks

“…Methods employing long short-term memory (LSTM) units [13] are particularly powerful in areas such as speech recognition [14] and are nowadays applied in astronomy in different variants (e.g. [15,16]). Lately the development of combined convolutional and recurrent neural networks (CRNNs) have been explored to classify transient objects in optical astronomy (e.g.…”

fink, a new generation of broker for the LSST community

“…F will deliver classification, after only a few observations, in order to select promising candidates for further analysis and follow-up coordination (e.g. Möller & de Boissière 2019). This early identification is crucial to allow optimal distribution of follow-up efforts for further SNe studies, including spectroscopic typing of supernovae, as well as to improve training sets for photometric classifiers (Ishida et al 2019b).…”

“…This challenging task requires algorithms that are able to characterise objects with only a handful of light-curve observations (e.g. Möller & de Boissière 2019;Muthukrishna et al 2019;Godines et al 2019;Jamal & Bloom 2020).…”