Abstract:Astronomical transients, such as supernovae and other rare stellar explosions, have been instrumental in some of the most significant discoveries in astronomy. New astronomical sky surveys will soon record unprecedented numbers of transients as sparsely and irregularly sampled multivariate time series. To improve our understanding of the physical mechanisms of transients and their progenitor systems, early-time measurements are necessary. Prioritizing the follow-up of transients based on their age along with their class is crucial for new surveys. To meet this demand, we present the first method of predicting the age of transients in real-time from multi-wavelength time-series observations. We build a Bayesian probabilistic recurrent neural network. Our method can accurately predict the age of a transient with robust uncertainties as soon as it is initially triggered by a survey telescope. This work will be essential for the advancement of our understanding of the numerous young transients being detected by ongoing and upcoming astronomical surveys.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: How to predict the age of astronomical transient sources (such as supernovae and other rare stellar explosions) from real - time multivariate time - series data. Specifically, the research aims to develop a method that can quickly and accurately predict the age of a transient source when it is first triggered by a survey telescope, with robust uncertainty estimates. ### Problem Background With the launch of new astronomical survey projects (such as the Legacy Survey of Space and Time (LSST) of the Vera Rubin Observatory), an unprecedented number of transient sources will be recorded every day. The amount of data for these transient sources is huge, but the global resources for follow - up observations are limited, so only a small fraction of them can be observed in detail spectroscopically or in multiple bands. For a better understanding of the physical mechanisms of transient sources and their progenitor systems, early observations are crucial. However, most of the current methods for automatically identifying transient sources require complete phase - coverage data, which limits the scientific questions that can be answered. ### Research Objectives This paper proposes a method based on Bayesian probabilistic Recurrent Neural Networks (RNNs) that can predict the age of transient sources in real - time from multi - wavelength time - series observational data. This method can accurately predict the age of a transient source within a short time (ideally within a few days) after it is first triggered, thus helping to prioritize young transient sources for follow - up observations and facilitating rapid follow - up and detailed observations of their progenitor systems. ### Method Overview 1. **Dataset**: The authors used multi - channel light - curve data of 805 transient sources observed by the Zwicky Transient Facility (ZTF) and the Transiting Exoplanet Survey Satellite (TESS). 2. **Model Construction**: - Two independent RNN models were constructed, one for predicting the first - light time \( t_0 \) and the other for predicting the time of maximum luminosity \( t_{\text{max}} \). - The model input is a multi - band light - curve matrix \( X_s(t \leq T) \), and the output is a probabilistic prediction \( Y_w^s \) of the transient - source age, parameterized as a normal distribution with a predicted mean \( \hat{y}_w^s \) and a predicted uncertainty \( \hat{\sigma}_w^s \). - Two unidirectional RNN layers of 50 units were constructed using Gated Recurrent Units (GRUs), and finally the predicted mean and standard deviation were output through a time - distributed fully - connected probabilistic layer. 3. **Uncertainty Quantification**: - The uncertainty of the model prediction was quantified by the Monte Carlo (MC) dropout sampling method, posterior samples were generated, and the marginal predicted mean \( \hat{y}_s \) and the predicted uncertainty \( \hat{\sigma}_s \) were calculated. 4. **Performance Evaluation**: - The performance of the model was verified on the test set, and the results showed that the mean absolute errors after the trigger moment were \( \text{MAE}_{t_0} = 2.1 \) days and \( \text{MAE}_{t_{\text{max}}} = 4.2 \) days respectively. - As more light - curve data are obtained, the prediction accuracy is significantly improved, especially for the prediction of the first - light time, which can reach a high accuracy at the trigger moment. ### Conclusions This research presents the first method that can predict the age of transient sources under real - time conditions, which is applicable to large - scale astronomical survey projects such as LSST. This method not only improves the identification efficiency of young transient sources but also provides strong support for further in - depth research on their progenitor systems. Future work will further optimize the model architecture, explore the possibility of using Transformers to replace RNNs, and apply it to actual transient - source survey projects.

Predicting the Age of Astronomical Transients from Real-Time Multivariate Time Series

Real-time Detection of Anomalies in Multivariate Time Series of Astronomical Data

Real-Time Detection of Anomalies in Large-Scale Transient Surveys

Detecting optical transients using artificial neural networks and reference images from different surveys

First Impressions: Early-Time Classification of Supernovae using Host Galaxy Information and Shallow Learning

Classifying Image Sequences of Astronomical Transients with Deep Neural Networks

Transient Classification in low SNR Gravitational Wave data using Deep Learning

A Classifier-Based Approach to Multi-Class Anomaly Detection for Astronomical Transients

Automated Real-Time Classification and Decision Making in Massive Data Streams from Synoptic Sky Surveys

Real-Time Data Mining of Massive Data Streams from Synoptic Sky Surveys

A transient search using combined human and machine classifications

Optical Transient Object Classification in Wide Field Small Aperture Telescopes with Neural Networks

Transient-optimised real-bogus classification with Bayesian Convolutional Neural Networks -- sifting the GOTO candidate stream

Automating Discovery and Classification of Transients and Variable Stars in the Synoptic Survey Era

RAPID: Early Classification of Explosive Transients using Deep Learning

Real-time, Value-driven Data Augmentation in the Era of LSST

Nonparametric Transient Classification using Adaptive Wavelets

Understanding of the properties of neural network approaches for transient light curve approximations

Towards Rate Estimation for Transient Surveys I: Assessing Transient Detectability and Volume Sensitivity for iPTF

The Zwicky Transient Facility Bright Transient Survey. II. A Public Statistical Sample for Exploring Supernova Demographics

Real-Time Detection of Unmodelled Gravitational-Wave Transients Using Convolutional Neural Networks