Despite the utility of neural networks (NNs) for astronomical time-series classification, the proliferation of learning architectures applied to diverse datasets has thus far hampered a direct intercomparison of different approaches. Here we perform the first comprehensive study of variants of NN-based learning and inference for astronomical time-series, aiming to provide the community with an overview on relative performance and, hopefully, a set of best-in-class choices for practical implementations. In both supervised and self-supervised contexts, we study the effects of different time-series-compatible layer choices, namely the dilated temporal convolutional neural network (dTCNs), Long-Short Term Memory (LSTM) NNs, Gated Recurrent Units (GRUs) and temporal convolutional NNs (tCNNs). We also study the efficacy and performance of encoder-decoder (i.e., autoencoder) networks compared to direct classification networks, different pathways to include auxiliary (non-time-series) metadata, and different approaches to incorporate multi-passband data (i.e., multiple time-series per source). Performance---applied to a sample of 17,604 variable stars from the MACHO survey across 10 imbalanced classes---is measured in training convergence time, classification accuracy, reconstruction error, and generated latent variables. We find that networks with Recurrent NN (RNNs) generally outperform dTCNs and, in many scenarios, yield to similar accuracy as tCNNs. In learning time and memory requirements, convolution-based layers are more performant. We conclude by discussing the advantages and limitations of deep architectures for variable star classification, with a particular eye towards next-generation surveys such as LSST, WFIRST and ZTF2.
With an ever-increasing amount of astronomical data being collected, manual classification has become obsolete; and machine learning is the only way forward. Keeping this in mind, the Large Synoptic Survey Telescope (LSST) Team hosted the Photometric LSST Astronomical Time-Series Classification Challenge (PLAsTiCC) in 2018. The aim of this challenge was to develop models that accurately classify astronomical sources into different classes, scaling from a limited training set to a large test set. In this text, we report our results of experimenting with Bidirectional Gated Recurrent Unit (GRU) …
Website with interactive visualization of multivariate astronomical time series
Yandex LLC（Yandex公司）；Max-Planck-Institut für Astronomie（马克斯·普朗克天文学研究所）；Heidelberg University（海德堡大学）；Sternberg Astronomical Institute, Moscow M.V. Lomonosov State University（莫斯科罗蒙诺索夫国立大学斯特恩伯格天文研究所）；Université Paris Cité, CNRS(/IN2P3), Astroparticule et Cosmologie（巴黎西岱大学，法国国家科学研究中心/法国核能与替代能源委员会，亚原子粒子与宇宙学实验室）；Center for Astrophysics — Harvard and Smithsonian（哈佛-史密森尼天体物理中心）
Context. K-means is a clustering algorithm that has been used to classify large datasets in astronomical databases. It is an unsupervised method, able to cope very different types of problems. Aims. We check whether a variant of the algorithm called single-pass k-means can be used as a fast alternative to the traditional k-means. Methods. The execution time of the two algorithms are compared when classifying subsets drawn from the SDSS-DR7 catalog of galaxy spectra. Results. Single-pass k-means turn out to be between 20 % and 40 % faster than k-means and provide statistically equivalent classi…
The exploitation of present and future synoptic (multi-band and multi-epoch) surveys requires an extensive use of automatic methods for data processing and data interpretation. In this work, using data extracted from the Catalina Real Time Transient Survey (CRTS), we investigate the classification performance of some well tested methods: Random Forest, MLPQNA (Multi Layer Perceptron with Quasi Newton Algorithm) and K-Nearest Neighbors, paying special attention to the feature selection phase. In order to do so, several classification experiments were performed. Namely: identification of catacly…
Upcoming astronomical surveys such as the Large Synoptic Survey Telescope (LSST) will rely on photometric classification to identify the majority of the transients and variables that they discover. We present a set of techniques for photometric classification that can be applied even when the training set of spectroscopically-confirmed objects is heavily biased towards bright, low-redshift objects. Using Gaussian process regression to model arbitrary light curves in all bands simultaneously, we "augment" the training set by generating new versions of the original light curves covering a range …
Classification is valuable and necessary in spectral analysis, especially for data-driven mining. Along with the rapid development of spectral surveys, a variety of classification techniques have been successfully applied to astronomical data processing. However, it is difficult to select an appropriate classification method in practical scenarios due to the different algorithmic ideas and data characteristics. Here, we present the second work in the data mining series - a review of spectral classification techniques. This work also consists of three parts: a systematic overview of current lit…
This paper describes the VARTOOLS program, which is an open-source command-line utility, written in C, for analyzing astronomical time-series data, especially light curves. The program provides a general-purpose set of tools for processing light curves including signal identification, filtering, light curve manipulation, time conversions, and modeling and simulating light curves. Some of the routines implemented include the Generalized Lomb-Scargle periodogram, the Box-Least Squares transit search routine, the Analysis of Variance periodogram, the Discrete Fourier Transform including the CLEAN…
Time series data of celestial objects are commonly used to study valuable and unexpected objects such as extrasolar planets and supernova in time domain astronomy. Due to the rapid growth of data volume, traditional manual methods are becoming extremely hard and infeasible for continuously analyzing accumulated observation data. To meet such demands, we designed and implemented a special tool named AstroCatR that can efficiently and flexibly reconstruct time series data from large-scale astronomical catalogues. AstroCatR can load original catalogue data from Flexible Image Transport System (FI…
The Zwicky Transient Facility (ZTF), a state-of-the-art optical robotic sky survey, registers on the order of a million transient events - such as supernova explosions, changes in brightness of variable sources, or moving object detections - every clear night, and generates associated real-time alerts. We present Alert-Classifying Artificial Intelligence (ACAI), an open-source deep-learning framework for the phenomenological classification of ZTF alerts. ACAI uses a set of five binary classifiers to characterize objects which, in combination with the auxiliary/contextual event information avai…
We present a classification-based approach to identify quasi-stellar radio sources (quasars) in the Sloan Digital Sky Survey and evaluate its performance on a manually labeled training set. While reasonable results can already be obtained via approaches working only on photometric data, our experiments indicate that simple but problem-specific features extracted from spectroscopic data can significantly improve the classification performance. Since our approach works orthogonal to existing classification schemes used for building the spectroscopic catalogs, our classification results are well …
Studies in Astronomical Time Series Analysis: The Double Lomb-Scargle Periodogram and Super Resolution
NASA Ames Research Center（NASA 艾姆斯研究中心）；Julius-Maximilians-Universität Würzburg（维尔茨堡大学）；Stanford University（斯坦福大学）；Institut de Física d’Altes Energies (IFAE)（高能物理研究所 (IFAE)）；The Barcelona Institute of Science and Technology (BIST)（巴塞罗那科学技术研究院 (BIST)）
This paper reviews the Stochastic Recurrent Neural Network (SRNN) as applied to the light curves of Active Galactic Nuclei by Sheng et al. (2022). Astronomical data have inherent limitations arising from telescope capabilities, cadence strategies, inevitable observing weather conditions, and current understanding of celestial objects. When applying machine learning methods, it is vital to understand the effects of data limitations on our analysis and ability to make inferences. We take Sheng et al. (2022) as a case study, and illustrate the problems and limitations encountered in implementing …
The Photometric LSST Astronomical Time Series Classification Challenge (PLAsTiCC) is an open data challenge to classify simulated astronomical time-series data in preparation for observations from the Large Synoptic Survey Telescope (LSST), which will achieve first light in 2019 and commence its 10-year main survey in 2022. LSST will revolutionize our understanding of the changing sky, discovering and measuring millions of time-varying objects. In this challenge, we pose the question: how well can we classify objects in the sky that vary in brightness from simulated LSST time-series data, with…
Detection of a signal hidden by noise within a time series is an important problem in many astronomical searches, i.e. for light curves containing the contributions of periodic/semi-periodic components due to rotating objects and all other astrophysical time-dependent phenomena. One of the most popular tools for use in such studies is the "periodogram", whose use in an astronomical context is often not trivial. The "optimal" statistical properties of the periodogram are lost in the case of irregular sampling of signals, which is a common situation in astronomical experiments. Parts of these pr…
This paper introduces the multiband periodogram, a general extension of the well-known Lomb-Scargle approach for detecting periodic signals in time-domain data. In addition to advantages of the Lomb-Scargle method such as treatment of non-uniform sampling and heteroscedastic errors, the multiband periodogram significantly improves period finding for randomly sampled multiband light curves (e.g., Pan-STARRS, DES and LSST). The light curves in each band are modeled as arbitrary truncated Fourier series, with the period and phase shared across all bands. The key aspect is the use of Tikhonov regu…
The last two decades have seen a major expansion in the availability, size, and precision of time-domain datasets in astronomy. Owing to their unique combination of flexibility, mathematical simplicity and comparative robustness, Gaussian Processes (GPs) have emerged recently as the solution of choice to model stochastic signals in such datasets. In this review we provide a brief introduction to the emergence of GPs in astronomy, present the underlying mathematical theory, and give practical advice considering the key modelling choices involved in GP regression. We then review applications of …
"exoplanet" is a toolkit for probabilistic modeling of astronomical time series data, with a focus on observations of exoplanets, using PyMC3 (Salvatier et al., 2016). PyMC3 is a flexible and high-performance model-building language and inference engine that scales well to problems with a large number of parameters. "exoplanet" extends PyMC3's modeling language to support many of the custom functions and probability distributions required when fitting exoplanet datasets or other astronomical time series. While it has been used for other applications, such as the study of stellar variability, t…
We present a new framework to detect various types of variable objects within massive astronomical time-series data. Assuming that the dominant population of objects is non-variable, we find outliers from this population by using a non-parametric Bayesian clustering algorithm based on an infinite GaussianMixtureModel (GMM) and the Dirichlet Process. The algorithm extracts information from a given dataset, which is described by six variability indices. The GMM uses those variability indices to recover clusters that are described by six-dimensional multivariate Gaussian distributions, allowing o…
The present work aims to validate the positions of solar filaments published in the Annals of Coimbra University Astronomical Observatory, currently the Geophysical and Astronomical Observatory of the University of Coimbra, corresponding to years 1929 to 1941. The published Stonyhurst positions were obtained by an original method devised in the early 20th century that used a spherical calculator instrument, a wood-made model of the sun. We used the digital images of the original spectroheliograms to measure the positions of the filaments, and heliographic coordinates were determinate with the …
In this letter, we propose a method for period estimation in light curves from periodic variable stars using correntropy. Light curves are astronomical time series of stellar brightness over time, and are characterized as being noisy and unevenly sampled. We propose to use slotted time lags in order to estimate correntropy directly from irregularly sampled time series. A new information theoretic metric is proposed for discriminating among the peaks of the correntropy spectral density. The slotted correntropy method outperformed slotted correlation, string length, VarTools (Lomb-Scargle period…
An automated, rapid classification of transient events detected in the modern synoptic sky surveys is essential for their scientific utility and effective follow-up using scarce resources. This presents some unusual challenges: the data are sparse, heterogeneous and incomplete; evolving in time; and most of the relevant information comes not from the data stream itself, but from a variety of archival data and contextual information (spatial, temporal, and multi-wavelength). We are exploring a variety of novel techniques, mostly Bayesian, to respond to these challenges, using the ongoing CRTS s…
In the last years, the need for automated real-time detection and classification of astronomical transients began to be more impelling. Better technologies involve a higher number of detected candidates and an automated classification will allow dealing with this amount of data, every night. The desired state-of-the-art in detection and classification will be presented in its key features and different practical approaches will be introduced, as well. Several ongoing and future surveys will be presented, showing the current situation of Time-Domain Astronomy, and eventually compared with the d…
We present a data-adaptive spectral method - Monte Carlo Singular Spectrum Analysis (MC-SSA) - and its modification to tackle astrophysical problems. Through numerical simulations we show the ability of the MC-SSA in dealing with $1/f^β$ power-law noise affected by photon counting statistics. Such noise process is simulated by a first-order autoregressive, AR(1) process corrupted by intrinsic Poisson noise. In doing so, we statistically estimate a basic stochastic variation of the source and the corresponding fluctuations due to the quantum nature of light. In addition, MC-SSA test retains its…
We propose a new sequential classification model for astronomical objects based on a recurrent convolutional neural network (RCNN) which uses sequences of images as inputs. This approach avoids the computation of light curves or difference images. This is the first time that sequences of images are used directly for the classification of variable objects in astronomy. The second contribution of this work is the image simulation process. We generate synthetic image sequences that take into account the instrumental and observing conditions, obtaining a realistic, set of movies for each astronomi…
We present the Spectroscopic Classification of Astronomical Transients (SCAT) survey, which is dedicated to spectrophotometric observations of transient objects such as supernovae and tidal disruption events. SCAT uses the SuperNova Integral-Field Spectrograph (SNIFS) on the University of Hawai'i 2.2-meter (UH2.2m) telescope. SNIFS was designed specifically for accurate transient spectrophotometry, including absolute flux calibration and host-galaxy removal. We describe the data reduction and calibration pipeline including spectral extraction, telluric correction, atmospheric characterization,…
Astronomical image time series classification using CONVolutional attENTION (ConvEntion)
University of Montpellier（蒙彼利埃大学）；University of Nimes（尼姆大学）；Aix Marseille Univ（艾克斯-马赛大学）；Paul Valéry University Montpellier 3（蒙彼利埃第三大学（保罗·瓦莱里大学））；INRAE（法国国家农业、食品与环境研究院）；CIRAD（法国国际农业研究合作中心）；CNRS（法国国家科学研究中心）
Detection, Instance Segmentation, and Classification for Astronomical Surveys with Deep Learning (DeepDISC): Detectron2 Implementation and Demonstration with Hyper Suprime-Cam Data
Universidad de Chile（智利大学）；Harvard University（哈佛大学）；Millennium Institute of Astrophysics（千年天体物理研究所）
Due to the latest advances in technology, telescopes with significant sky coverage will produce millions of astronomical alerts per night that must be classified both rapidly and automatically. Currently, classification consists of supervised machine learning algorithms whose performance is limited by the number of existing annotations of astronomical objects and their highly imbalanced class distributions. In this work, we propose a data augmentation methodology based on Generative Adversarial Networks (GANs) to generate a variety of synthetic light curves from variable stars. Our novel contr…
ASTRONIRCAM is an infrared camera-spectrograph installed at the 2.5-meter telescope of the CMO SAI. The instrument is equipped with the HAWAII-2RG array. A bad pixels classification of the ASTRONIRCAM detector is proposed. The classification is based on histograms of the difference of consecutive non-destructive readouts of a flat field. Bad pixels are classified into 5 groups: hot (saturated on the first readout), warm (the signal accumulation rate is above the mean value by more than 5 standard deviations), cold (the rate is under the mean value by more than 5 standard deviations), dead (no …
Wide field small aperture telescopes are widely used for optical transient observations. Detection and classification of astronomical targets in observed images are the most important and basic step. In this paper, we propose an astronomical targets detection and classification framework based on deep neural networks. Our framework adopts the concept of the Faster R-CNN and uses a modified Resnet-50 as backbone network and a Feature Pyramid Network to extract features from images of different astronomical targets. To increase the generalization ability of our framework, we use both simulated a…
Fast Inference on Astronomical Time Series with Trans-Dimensional Flow Matching Posterior Estimation
Anton Pannekoek Institute University of Amsterdam（阿姆斯特丹大学安东·潘内科克研究所）；Informatics Institute University of Amsterdam（阿姆斯特丹大学信息学研究所）；GRAPPA Institute Institute for Theoretical Physics Amsterdam University of Amsterdam（阿姆斯特丹大学理论物理研究所 GRAPPA 研究所）
Universidad De Concepcion（康塞普西翁大学）；Millennium Institute of Astrophysics（千年天体物理研究所）；Harvard University（哈佛大学）；University College London（伦敦大学学院）；Universidad de Chile（智利大学）
In the new era of very large telescopes, where data is crucial to expand scientific knowledge, we have witnessed many deep learning applications for the automatic classification of lightcurves. Recurrent neural networks (RNNs) are one of the models used for these applications, and the LSTM unit stands out for being an excellent choice for the representation of long time series. In general, RNNs assume observations at discrete times, which may not suit the irregular sampling of lightcurves. A traditional technique to address irregular sequences consists of adding the sampling time to the networ…
Astronomy has entered the multi-messenger data era and Machine Learning has found widespread use in a large variety of applications. The exploitation of synoptic (multi-band and multi-epoch) surveys, like LSST (Legacy Survey of Space and Time), requires an extensive use of automatic methods for data processing and interpretation. With data volumes in the petabyte domain, the discrimination of time-critical information has already exceeded the capabilities of human operators and crowds of scientists have extreme difficulty to manage such amounts of data in multi-dimensional domains. This work i…
AstronomicAL is a human-in-the-loop interactive labelling and training dashboard that allows users to create reliable datasets and robust classifiers using active learning. This technique prioritises data that offer high information gain, leading to improved performance using substantially less data. The system allows users to visualise and integrate data from different sources and deal with incorrect or missing labels and imbalanced class sizes. AstronomicAL enables experts to visualise domain-specific plots and key information relating both to broader context and details of a point of intere…
The production of photometric light curves from astronomical images is a very time-consuming task. Larger data sets improve the resolution of the light curve, however, the time requirement scales with data volume. The data analysis is often made more difficult by factors such as a lack of suitable calibration sources and the need to correct for variations in observing conditions from one image to another. Often these variations are unpredictable and corrections are based on experience and intuition. The High Efficiency Image Detection & Identification (HEIDI) pipeline software rapidly processe…
A broad and widely used class of stationary, linear, additive time series models can have statistical properties which many authors have asserted imply that the underlying process must be non-linear, non-stationary, multiplicative, or inconsistent with shot noise. This result is demonstrated with exact and numerical evaluation of the model flux distribution function and dependence of flux standard deviation on mean flux (here and in the literature called the \emph{rms-flux relation}). These models can: (1) exhibit normal, log-normal or other flux distributions; (2) show linear or slightly non-…
We present a Bayesian method for the identification and classification of objects from sets of astronomical catalogs, given a predefined classification scheme. Identification refers here to the association of entries in different catalogs to a single object, and classification refers to the matching of the associated data set to a model selected from a set of parametrized models of different complexity. By the virtue of Bayes' theorem, we can combine both tasks in an efficient way, which allows a largely automated and still reliable way to generate classified astronomical catalogs. A problem t…
Celestial objects exhibit a wide range of variability in brightness at different wavebands. Surprisingly, the most common methods for characterizing time series in statistics -- parametric autoregressive modeling -- is rarely used to interpret astronomical light curves. We review standard ARMA, ARIMA and ARFIMA (autoregressive moving average fractionally integrated) models that treat short-memory autocorrelation, long-memory $1/f^α$ `red noise', and nonstationary trends. Though designed for evenly spaced time series, moderately irregular cadences can be treated as evenly-spaced time series wit…
Most general-purpose classification methods, such as support-vector machine (SVM) and random forest (RF), fail to account for an unusual characteristic of astronomical data: known measurement error uncertainties. In astronomical data, this information is often given in the data but discarded because popular machine learning classifiers cannot incorporate it. We propose a simulation-based approach that incorporates heteroscedastic measurement error into existing classification method to better quantify uncertainty in classification. The proposed method first simulates perturbed realizations of …
Simulation-Based Pretraining and Domain Adaptation for Astronomical Time Series with Minimal Labeled Data
Massachusetts Institute of Technology, Cambridge, MA 02139, USA；Irvington High School, Fremont, CA 94538, USA；Center for Astrophysics, Harvard \& Smithsonian, Cambridge, MA 02138, USA
Classification of transient and variable light curves is an essential step in using astronomical observations to develop an understanding of their underlying physical processes. However, upcoming deep photometric surveys, including the Large Synoptic Survey Telescope (LSST), will produce a deluge of low signal-to-noise data for which traditional labeling procedures are inappropriate. Probabilistic classification is more appropriate for the data but are incompatible with the traditional metrics used on deterministic classifications. Furthermore, large survey collaborations intend to use these c…
StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars
Department of Computer Science, Northwestern University（计算机科学系，西北大学）；Center for Foundation Models and Generative AI, Northwestern University（基础模型与生成AI中心，西北大学）；NSF – Simons AI Institute for the Sky (SkAI)（国家科学基金会-斯隆人工智能天文研究所（SkAI））；Department of Physics and Astronomy, Northwestern University（物理与天文学系，西北大学）；Center for Interdisciplinary Exploration and Research in Astrophysics (CIERA)（天文学跨学科探索与研究中心（CIERA））；Department of Statistics and Data Science, Northwestern University（统计与数据科学系，西北大学）
