WWW.DISSERTATION.XLIBX.INFO
FREE ELECTRONIC LIBRARY - Dissertations, online materials
 
<< HOME
CONTACTS



Pages:   || 2 |

«Towards Real-time Classification of Astronomical Transients A. Mahabal*, S. G. Djorgovski*, R. Williams*, A. Drake*, C. Donalek*, M. Graham*, B. ...»

-- [ Page 1 ] --

Towards Real-time Classification of

Astronomical Transients

A. Mahabal*, S. G. Djorgovski*, R. Williams*, A. Drake*, C. Donalek*, M.

Graham*, B. Moghaddam''', M. Turmon''', J. Jewell''', A. Khosla* and B.

Hensley*

'California Institute of Technology, MC105-24, 1200 E. California BL, Pasadena, CA 91125, USA

^Jet Propulsion Laboratory, 4800 Oak Grove Dr., Pasadena, CA 91109, USA

Abstract. Exploration of time domain is now a vibrant area of research in astronomy, driven

by the advent of digital synoptic sky surveys. While panoramic surveys can detect variable or transient events, typically some follow-up observations are needed; for short-lived phenomena, a rapid response is essential. Ability to automatically classify and prioritize transient events for follow-up studies becomes critical as the data rates increase. We have been developing such methods using the data streams from the Palomar-Quest survey, the Catalina Sky Survey and others, using the VOEventNet framework. The goal is to automatically classify transient events, using the new measurements, combined with archival data (previous and multi-wavelength measurements), and contextual information (e.g.. Galactic or ecliptic latitude, presence of a possible host galaxy nearby, etc.); and to iterate them dynamically as the follow-up data come in (e.g., light curves or colors).

We have been investigating Bayesian methodologies for classification, as well as discriminated follow-up to optimize the use of available resources, including Naive Bayesian approach, and the non-parametric Gaussian process regression. We will also be deploying variants of the traditional machine learning techniques such as Neural Nets and Support Vector Machines on datasets of reliably classified transients as they build up.

Keywords: Transients; Classification; Bayesian Techniques; Machine Learning PACS: 95.75.-z,95.80.+p,95.85.-e,95.85.Kr,98.70.-f

INTRODUCTION

Time domain astronomy has rapidly emerged as one of the more exciting areas of research in astronomy. It touches on a number of important scientific directions, ranging from exploration of the Solar System to cosmology. Besides the objects that move (e.g., asteroids, TNOs, KBOs), the types of transients we are likely to encounter include SNe (cosmological standard candles, as well as endpoints of stellar evolution), GRB orphan afterglows (which constrain the beaming models), variable stars of all sorts (probes of stellar astrophysics and Galactic structure), AGN (as a method of finding QSOs and constraining their fueling mechanisms and lifetimes), etc. There are classes of variable events which are expected or suspected to occur, but for which there is only a limited evidence in hand, e.g., tidal disruption of stars by otherwise quiescent supermassive black holes [1], breakout shocks of Type II SNe [2, 3], or mega-flares on otherwise normal, main-sequence stars [4], etc.

We have been exploring varibles and transients from the Palomar-Quest Sky Survey [12] ( h t t p : / / w w w. p a l q u e s t. o r g ) and the Catalina Sky Survey ( h t t p :

CP1082, Classification and Discovery in Large Astronomical Surveys, edited by C. A. L. Bailer-Jones © 2008 American Institute of Physics 978-0-7354-0613-l/08/$23.00 Downloaded 15 Dec 2008 to 131.215.225.9. Redistribution subject to AIP license or copyright; see http://proceedings.aip.org/proceedings/cpcr.jsp / /www. I p l. a r i z o n a. e d u / c s s ) in real-time and announcing those via VOEventNet ( h t t p : / / v o e v e n t n e t. c a l t e c h. edu). Besides the optical surveys (e.g. PTF, LSST, Pan-STARRS) there are many at other wavelengths which will find transients and whose science can be enhanced by real-time classification of these transients (e.g. Fermi, SKA, LOFAR, LISA to name a few).

From targeted observing of small samples of a particular type of variable objects or phenomena, the field has been moving towards a systematic exploration of larger areas with a better time sampling and understanding of finer details of these phenomena. Rapid follow-up is essential for proper understanding and scientific exploitation of the events varying on a short time scale, or unusual classes of objects. Many of these objects do not have counterparts in archival image surveys like DSS, DPOSS, 2MASS etc. making the discovery data points and any follow-up the sole data to go on.

An illustrative example was OT SNF143933+054631 discovered in the Point-andStare data from Palomar-Quest by the LBNL SNF using archival comparison images.

The initial SNIFs spectrum was highly unusual for a SN, and defied classification. Using follow-up imaging with the Palomar 60-inch telescope and a spectrum at Keck we were finally able to understand the nature of this peculiar SN [5]. Another case was SN20061t, which turned out to be a rare SN of type lb [6, 7]. We may have discovered a class of SN associated with faint dwarf galaxies in the process of looking for transients [8].

SN and GRB are of course not the only transients for which unusual classes are found. CSS080924:233423+391423 seemed to be a simple flaring object until foUow-up imaging revealed that it had persisted 24 hours later [9] and Fig. 1). A Palomar 200-inch spectrum revealed numerous emission lines at zero redshift typical of Galactic dwarf novae. The atypically large variations at discovery remain unexplained [10]. Yet another example is CSS080928:160837+041626, a possible high amplitude ( - 5 mags), long period variable, but with colors unlike one [11].





The need for quick reporting and follow-up has resulted in (1) the emergence of computer networks and protocols for collecting and distributing streams of interesting events from large surveys - the VOEventNet system which serves events from a multitude of streams including Palomar-Quest survey and the Catalina Sky Survey is a pertinent example [12], and (2) a number of robotic telescopes which can turn to a target very quickly and provide crucial data for the classification of the events.

Here we describe the current status of the real-time event classification effort. The endeavour is clearly applicable to other synoptic sky surveys. As the event streams from synoptic sky surveys such as LSST and SKA increase, real-time classification will become even more crucial as there will not be enough facilities for follow-up observations making real-time classification a key enabler of future synoptic astronomy.

A key difficulty of real-time classification of transients is the general lack of available information initially available. A transient detected by an increase in brightness is often missing in archival sky surveys and may have just a couple of relatively closely spaced observations in a couple of epochs to go by. Machine Learning methodologies including Support Vector Machines (SVMs), Artificial Neural Networks (ANNs) can be used as also Bayesian classifiers including Naive Bayesian algorithms and Gaussian Process regression [13].

Downloaded 15 Dec 2008 to 131.215.225.9. Redistribution subject to AIP license or copyright; see http://proceedings.aip.org/proceedings/cpcr.jsp 1..^,-... _ '^, * * ^ * • FIGURE 1. CSS080924:233423+391423 was classified as a transient based on its flaring within the CSS images taken minutes apart. Not found in any archival surveys except for the DSS N plate, it was still bright the next day. A spectrum later revealed it to be a dwarf nova, but its rapid variation remains unexplained. The three images on the left are individual CSS images taken on Sep 24 2008 UT, and the one on the right is a co-added baseline image, also from CSS.

METHODOLOGY

Bayesian Event Classification Given a small number of observations, it is not generally possible to unambiguously classify transients. The best approach in such a case is to calculate the probabilities of the object belonging to different classes of transients and then using objective criteria to determine if the probability for the class of interest is high enough. There will also be cases when none of the known classes is a good fit. This type may perhaps turn out to be most interesting, with the transient being a possible example of a new type of astronomical object or phenomenon.

We have described such a probabilistic method utilising Naive Bayes method in [14]. Briefly, the method involves building priors of different types of objects with a large number of features. The object to be classified has a feature vector which is decomposed into several independent blocks based on which class is being considered.

While this does not allow an exact membership to be determined, obtaining approximate probabilities while circumventing the curse of dimensionality as well as coping with several missing values makes this a powerful method. See also [15, 16].

Machine Learning in Classification

The machine learning, using ANNs/SVMs is more useful for dealing with variables where more prior data is available [14]. These too can deal with incomplete information, at least partially, by training several sets of quasi-independent classifiers and invoking the appropriate ones depending on what information is available (magnitudes, colors, shape parameters etc.).

Another way we have started to use these techniques is to classify (and eliminate) artifacts in real-time as data from telescopes are being processed. Many artifacts, based on some features present in their signature and the fact that they may not correspond to an object in the fiducial images, can get initially flagged as transients. This technique Downloaded 15 Dec 2008 to 131.215.225.9. Redistribution subject to AIP license or copyright; see http://proceedings.aip.org/proceedings/cpcr.jsp helps remove them. More details can be found in [17]. Similarly a very pertinent issue is star-galaxy classification from multi-epoch data. Due to varying conditions (seeing, airmass, extinction, filter) it is not possible to reproduce the same class in each individual epoch. Combining the disparate data can be ideally carried out using ANNs [17].

Feedback Incorporation Feedback incorporation [14] is an important step since every bit of additional information can potentially vindicate or contradict the original classification (or at least revise probabilities). In addition, the priors should also be updated so that classification in future has access to the new information for the corresponding classes. This can be carried out using either Expectation-Maximization algorithm [18] or kernel density estimation [19]. The unknown parameters in these approaches can be determined using known physical parameters for given classes (e.g. SNs do not normally increase in brightness once they start fading, or RR-Lyrae do not have their own host galaxies etc.) or additional observations as they become available or labelling done by experts. This final bit plays an important role in semantically connected portfolios built for each transient. Such transient portfolios are structured yet flexible annotation mechanisms including images and spectra besides comments. The collection of portfolios can be indexed for ease of searching and execution of need-based services [20].

Follow-up Prioritization Engine Given the paucity of follow-up resources in the era of very large volume event streams, matching transients to follow with such resources may well be the most crucial step in advancing our knowledge about rare classes of quickly fading objects. It will also be crucial in breaking the ambiguity between two possible classes. An informationtheoretic approach leading to the reduction of final entropy is the best choice [21, 14].

Gaussian Process Regression Given a small number of observations in a single band at different epochs, Gaussian Process Regression (GP) can be an effctive in determining class membership. In this process priors are built from lightcurves of objects belonging to a class of objects. For instance, consider just two epochs of a newly discovered transient (Fig. 2). One can compare the two points against different parts of a lightcurve in order to determine if there is a possible fit and if so at what stage of evolution (e.g. periodicity). The lower panel shows such a comparison with log marginal likelihood on the Y-axis. The more observed points one has, the more stringent results one can obtain. Fig 3 shows different possibilities involving 3 points (i.e. 2 dt and corresponding dm). For a given class of objects one can visualize a surface made of a grid of dt and dm values which can then be used to determine class membership. However, owing to the variation in objects belonging to a class, such a surface is somewhat fuzzy.

We have been conducting preliminary investigations using SN and Mira light curves and will be extending it to other light curves. The results so far are encouraging.

VOEventNet

VOEventNet ( h t t p : / / v o e v e n t n e t. c a l t e c h. edu)federates streams of astronomical events such that humans as well as robotic telescopes can subscribe to the events. The events are available in real-time in a standard format. The different streams allow subscribers to choose events of their interest (e.g. SN, GRB, asteroids etc.). Google Sky serves VOEventNet events under their "Current Sky Events" with links to related technical and astronomical information. A color scheme allows more recent events to stand out. AAVSO streams are also expected to be available under VOEventNet soon.

Downloaded 15 Dec 2008 to 131.215.225.9. Redistribution subject to AIP license or copyright; see http://proceedings.aip.org/proceedings/cpcr.jsp FIGURE 3. Given an object type one can build a surface indicating what likely dm& at what dt& one is likely to encounter. One can then compare any such pair against such a surface to see if the pair of points come from an object of that type. In practise we of course do not have such a crisp surface.

The reecntly concluded Palomar-Quest survey and the ongoing Catalina Sky Survey utilize VOEventNet to serve variables and transients in real-time. This allowed quick follow-up in many cases and has resulted in datasets which would otherwise be difficult to compile. For instance, during the first six months of its real-time observations, CSS found 350 transients [8]. 240 of these were SN or CV and the remaining included AGNs, high proper motion stars, highly variable stars as well as blazars and transients of an unknown nature.



Pages:   || 2 |


Similar works:

«HOW ASTROPHYSICAL MEAN FIELD DYNAMOS CAN CIRCUMVENT EXISTING QUENCHING CONSTRAINTS Eric G. Blackman, Department of Physics & Astronomy University of Rochester, Rochester NY 14627 and George B. Field, Harvard-Smithsonian Center for Astrophysics (CFA) 60 Garden St., Cambridge MA, 02139, USA Abstract Mean field dynamo theory is a leading candidate to explain the large scale magnetic flux in galaxies and stars. However, controversy arises over the extent of premature quenching by the backreaction...»

«Models of Gemini Surfactants∗ Haim Diamant1 and David Andelman2 1 School of Chemistry and 2 School of Physics & Astronomy, Beverly and Raymond Sackler Faculty of Exact Sciences, Tel Aviv University, Tel Aviv 69978, Israel arXiv:cond-mat/0302184v1 [cond-mat.soft] 10 Feb 2003 I. INTRODUCTION Gemini surfactants are composed of two monomeric surfactant molecules linked by a spacer chain. They constitute a new class of amphiphilic molecules having its own distinct behavior. Since their first...»

«1 ™ Cherilynn Morrow and Michael Zawaski On the Astronomical Meaning of the Day, Year, and Seasons © Dr. Cherilynn A. Morrow and Michael Zawaski Kinesthetic Astronomy camorrow@colorado.edu and zawaski@colorado.edu Sky Time: 10 August 2004 Dear Educator, Here we present a significant upgrade to the first in a series of innovative, experiential lessons we call Kinesthetic Astronomy. The Sky Time lesson reconnects students with the astronomical meaning of the day, year, and seasons. Like all...»

«Chemical Abundances of Bright Giants in the Mildly Metal-Poor Globular Cluster M4 Inese I. Ivans∗ Department of Astronomy and McDonald Observatory, University of Texas at Austin, USA Abstract: We present a chemical composition analysis of three dozen giant stars in the nearby “CN-bimodal” mildly metal-poor ([Fe/H] = -1.18) globular cluster M4. The analysis combined traditional spectroscopic abundance methods with modifications to the line-depth ratio technique pioneered by Gray (1994)....»

«ANALYSIS OF SUPERNOVA SPECTRA DAVID J. JEFFERY AND DAVID BRANCH Department of Physics and Astronomy University of Oklahoma Norman, Oklahoma 73019, U.S.A. Reference Jeffery, D. J., & Branch, D. 1990, in Jerusalem Winter School for Theoretical Physics, Vol. 6, Supernovae, ed. J.C. Wheeler, T. Piran, and S. Weinberg (Singapore: World Scientific), p. 149. ANALYSIS OF SUPERNOVA SPECTRA DAVID J. JEFFERY AND DAVID BRANCH Department of Physics and Astronomy University of Oklahoma, Norman, Oklahoma...»

«Chela-Flores, J. Jerse, G., Messerotti, M. And Tuniz, C. (2008) Astronomical and astrobiological imprints on the fossil records. A review. “From Fossils to Astrobiology”, Ed. J. Seckbach, Cellular Origins, Life in Extreme Habitats and Astrobiology, Springer, Dordrecht, The Netherlands, pp. 389-408.ASTRONOMICAL AND ASTROBIOLOGICAL IMPRINTS ON THE FOSSIL RECORDS. A REVIEW J. CHELA-FLORES 1,2, G. JERSE 3, M. MESSEROTTI 4,3 AND C. TUNIZ 1 The Abdus Salam ICTP, Strada Costiera 11, 34014 Trieste,...»

«Juan Bautista de la Salle, vida y obra. Hno. Juan Bautista De las Heras 1.PRESENTACIÓN Desde los primeros biógrafos hasta nuestros días, muchos han sido los libros que se han escrito en torno a Juan Bautista de La Salle. Este escrito no pretende agotar el tema. En todo caso, recordaremos los hitos más importantes de su vida, fijándonos, sobre todo, en aquellos acontecimientos que fueron configurando la fundación del Instituto de los Hermanos de las Escuelas Cristianas. También...»

«Published in Theoretical and Applied Climatology 73(3-4): 133-150, 2002 Perturbations to astronomical observations at the European Southern Observatory’s very large telescope site in Paranal, Chile: analyses of climatological causes Martin Beniston1, Paula Casals1, and Marc Sarazin2 1: Department of Geosciences, University of Fribourg, Switzerland 2: European Southern Observatory, Garching, Germany Abstract A study has been conducted to assess the reasons for a significant decrease in the...»

«Emissivity Statistics in Turbulent, Compressible MHD Flows and the Density-Velocity Correlation Alex Lazarian1, Dmitri Pogosyan2, Enrique V´zquez-Semadeni3, and B´rbara Pichardo4 a a Dept. of Astronomy, University of Wisconsin, Madison, USA Canadian Institute for Theoretical Astrophysics, University of Toronto, CANADA Instituto de Astronom´ UNAM, Campus Morelia, Apdo. Postal 3-72, Xangari, 58089, Morelia, Mich., ıa, MEXICO Instituto de Astronom´ UNAM, Apdo. Postal 70-264, M´xico D.F....»

«Photon-detectors R&D for future neutrino experiments T. Nakaya Kyoto University, Kyoto, Japan Abstract We review the recent developments of new photon-detectors available for the future neutrino experiments. A photon-detector with large photo-coverage is the key component of a giant Cherenkov detector which will be an essential part of the future long baseline neutrino oscillation experiment to explore the CP violation in the neutrino sector. A photon-detector with Þne pixels is also the key...»

«Digitization and Scientific Exploitation of the Italian and Vatican Astronomical Plate Archives Cesare Barbieri1, Carlo Blanco3, Beatrice Bucciarelli4, Regina Coluzzi5, Andrea Di Paola5, Luciano Lanteri4, Gian Luca Li Causi5, Ettore Marilli10, Piero Massimino10, Vincenzo Mezzalira1, Stefano Mottola7, Roberto Nesci6, Alessandro Omizzolo2, Fernando Pedichini5, Francesca Rampazzi1, Corinne Rossi6, Ruggero Stagni1, Milcho Tsvetkov8, Roberto Viotti9 Department of Astronomy, University of Padova...»

«4-H 4-H Career Explorations Conference exploring Cornell and careers 2015 Program Facts What’s Inside Quick Details Quick Details General Information Dates: June 30-July 2, 2015 General 4-H Career Exploration Program Policies. 2 Location: Cornell University Preparing for the Conference Arrival: 10 AM – 12:30 PM on June 30 While you’re on Campus Departure: 2:30 PM on July 2 Program Descriptions Cost: $250 per person Focus for Teens Registration: Opens March 6, closes June 1 Astronomy...»





 
<<  HOME   |    CONTACTS
2016 www.dissertation.xlibx.info - Dissertations, online materials

Materials of this site are available for review, all rights belong to their respective owners.
If you do not agree with the fact that your material is placed on this site, please, email us, we will within 1-2 business days delete him.