arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1901.06384astro-ph.IM
收藏

SuperNNova:基于贝叶斯神经网络的超新星分类开源框架

SuperNNova: an open-source framework for Bayesian, Neural Network based supernova classification

Anais Möller, Thibault de Boissière

摘要:

我们介绍SuperNNova,这是一个开源的超新星光度分类框架,它利用了深度神经网络的最新进展。我们的核心算法是一个递归神经网络(RNN),它只使用光度信息进行训练来分类光曲线。附加的信息,如主星系的红移可以被纳入,以提高性能。我们评估我们的框架使用现实的超新星模拟,包括调查检测。我们表明,我们的方法,Ia型与非Ia超新星分类问题,达到精度大于96.92 +- 0.09没有任何红移的信息和高达99.55 +- 0.06时,红移,光度或光谱,是可用的。此外,我们还表明,我们的方法在不完整光变曲线的分类方面取得了前所未有的性能,在最大光前两天没有宿主星系红移(具有红移信息)的情况下,准确度达到>86.4 +- 0.1(>93.5 +- 0.8)。与以前的方法相比,不需要耗时的特征工程,我们证明了我们的方法可以以适度的计算预算扩展到非常大的数据集。此外,我们还研究了机器学习算法中经常被忽视的陷阱。我们发现,常用的算法遭受不良校准和过度自信的分布样本时,适用于超新星数据。我们设计了广泛的测试来估计分类器的鲁棒性,并在贝叶斯光下进行学习过程,从而更好地处理不确定性。我们研究了贝叶斯RNN对超新星Ia宇宙学的好处。我们的代码是开源的,可以在https://github.com/supernnova/SuperNNova上找到。

英文摘要:

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 on https://github.com/supernnova/SuperNNova.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑