arXiv subjects

Robert J. Brunner

Publications and source records attributed to Robert J. Brunner.

At least 19 recordsLinked to original sources

arXiv2020

Extended Isolation Forest

We present an extension to the model-free anomaly detection algorithm, Isolation Forest. This extension, named Extended Isolation Forest (EIF), resolves issues with assignment of anomaly score to given data points. We motivate the problem using heat maps for anomaly scores. These maps suffer from artifacts generated by the criteria for branching operation of the binary tree. We explain this problem in detail and demonstrate the mechanism by which it occurs visually. We then propose two different approaches for improving the situation. First we propose transforming the data randomly before creation of each tree, which results in averaging out the bias. Second, which is the preferred way, is to allow the slicing of the data to use hyperplanes with random slopes. This approach results in remedying the artifact seen in the anomaly score heat maps. We show that the robustness of the algorithm is much improved using this method by looking at the variance of scores of data points distributed along constant level sets. We report AUROC and AUPRC for our synthetic datasets, along with real-world benchmark datasets. We find no appreciable difference in the rate of convergence nor in computation time between the standard Isolation Forest and EIF.

cs.LG↗

arXiv2019

Unsupervised Star Galaxy Classification with Cascade Variational Auto-Encoder

The increasing amount of data in astronomy provides great challenges for machine learning research. Previously, supervised learning methods achieved satisfactory recognition accuracy for the star-galaxy classification task, based on manually labeled data set. In this work, we propose a novel unsupervised approach for the star-galaxy recognition task, namely Cascade Variational Auto-Encoder (CasVAE). Our empirical results show our method outperforms the baseline model in both accuracy and stability.

cs.LG↗

arXiv2016

Star-galaxy Classification Using Deep Convolutional Neural Networks

Most existing star-galaxy classifiers use the reduced summary information from catalogs, requiring careful feature extraction and selection. The latest advances in machine learning that use deep convolutional neural networks allow a machine to automatically learn the features directly from data, minimizing the need for input from human experts. We present a star-galaxy classification framework that uses deep convolutional neural networks (ConvNets) directly on the reduced, calibrated pixel values. Using data from the Sloan Digital Sky Survey (SDSS) and the Canada-France-Hawaii Telescope Lensing Survey (CFHTLenS), we demonstrate that ConvNets are able to produce accurate and well-calibrated probabilistic classifications that are competitive with conventional machine learning techniques. Future advances in deep learning may bring more success with current and forthcoming photometric surveys, such as the Dark Energy Survey (DES) and the Large Synoptic Survey Telescope (LSST), because deep neural networks require very little, manual feature engineering.

astro-ph.IM↗

arXiv2016

Teaching Data Science

We describe an introductory data science course, entitled Introduction to Data Science, offered at the University of Illinois at Urbana-Champaign. The course introduced general programming concepts by using the Python programming language with an emphasis on data preparation, processing, and presentation. The course had no prerequisites, and students were not expected to have any programming experience. This introductory course was designed to cover a wide range of topics, from the nature of data, to storage, to visualization, to probability and statistical analysis, to cloud and high performance computing, without becoming overly focused on any one subject. We conclude this article with a discussion of lessons learned and our plans to develop new data science courses.

stat.OT↗

arXiv2016

Machine Learning and Cosmological Simulations II: Hydrodynamical Simulations

We extend a machine learning (ML) framework presented previously to model galaxy formation and evolution in a hierarchical universe using N-body + hydrodynamical simulations. In this work, we show that ML is a promising technique to study galaxy formation in the backdrop of a hydrodynamical simulation. We use the Illustris Simulation to train and test various sophisticated machine learning algorithms. By using only essential dark matter halo physical properties and no merger history, our model predicts the gas mass, stellar mass, black hole mass, star formation rate, $g-r$ color, and stellar metallicity fairly robustly. Our results provide a unique and powerful phenomenological framework to explore the galaxy-halo connection that is built upon a solid hydrodynamical simulation. The promising reproduction of the listed galaxy properties demonstrably place ML as a promising and a significantly more computationally efficient tool to study small-scale structure formation. We find that ML mimics a full-blown hydrodynamical simulation surprisingly well in a computation time of mere minutes. The population of galaxies simulated by ML, while not numerically identical to Illustris, is statistically and physically robust and follows the same fundamental observational constraints. Machine learning offers an intriguing and promising technique to create quick mock galaxy catalogs in the future.

astro-ph.GA↗

arXiv2015

Creating updated, scientifically-calibrated mosaic images for the RC3 catalogue

The Third Reference Catalogue of Bright Galaxies (RC3) is a reasonably complete listing of 23,011 nearby, large, bright galaxies. By using the final imaging data release from the Sloan Digital Sky Survey, we generate scientifically-calibrated FITS mosaics by using the montage program for all SDSS imaging bands for all RC3 galaxies that lie within the survey footprint. We further combine the SDSS g, r, and i band FITS mosaics for these galaxies to create color-composite images by using the STIFF program. We generalized this software framework to make FITS mosaics and color-composite images for an arbitrary catalog and imaging data set. Due to positional inaccuracies inherent in the RC3 catalog, we employ a recursive algorithm in our mosaicking pipeline that first determines the correct location for each galaxy, and subsequently applies the mosaicking procedure. As an additional test of this new software pipeline and to obtain mosaic images of a larger sample of RC3 galaxies, we also applied this pipeline to photographic data taken by the Second Palomar Observatory Sky Survey with $B_J$, $R_F$, and $I_N$ plates. We publicly release all generated data, accessible via a web search form, and the software pipeline to enable others to make galaxy mosaics by using other catalogs or surveys.

astro-ph.IM↗

arXiv2015

Machine Learning and Cosmological Simulations I: Semi-Analytical Models

We present a new exploratory framework to model galaxy formation and evolution in a hierarchical universe by using machine learning (ML). Our motivations are two-fold: (1) presenting a new, promising technique to study galaxy formation, and (2) quantitatively analyzing the extent of the influence of dark matter halo properties on galaxies in the backdrop of semi-analytical models (SAMs). We use the influential Millennium Simulation and the corresponding Munich SAM to train and test various sophisticated machine learning algorithms (k-Nearest Neighbors, decision trees, random forests and extremely randomized trees). By using only essential dark matter halo physical properties for haloes of $M>10^{12} M_{\odot}$ and a partial merger tree, our model predicts the hot gas mass, cold gas mass, bulge mass, total stellar mass, black hole mass and cooling radius at z = 0 for each central galaxy in a dark matter halo for the Millennium run. Our results provide a unique and powerful phenomenological framework to explore the galaxy-halo connection that is built upon SAMs and demonstrably place ML as a promising and a computationally efficient tool to study small-scale structure formation.

astro-ph.GA↗

arXiv2015

A Hybrid Ensemble Learning Approach to Star-Galaxy Classification

There exist a variety of star-galaxy classification techniques, each with their own strengths and weaknesses. In this paper, we present a novel meta-classification framework that combines and fully exploits different techniques to produce a more robust star-galaxy classification. To demonstrate this hybrid, ensemble approach, we combine a purely morphological classifier, a supervised machine learning method based on random forest, an unsupervised machine learning method based on self-organizing maps, and a hierarchical Bayesian template fitting method. Using data from the CFHTLenS survey, we consider different scenarios: when a high-quality training set is available with spectroscopic labels from DEEP2, SDSS, VIPERS, and VVDS, and when the demographics of sources in a low-quality training set do not match the demographics of objects in the test data set. We demonstrate that our Bayesian combination technique improves the overall performance over any individual classification method in these scenarios. Thus, strategies that combine the predictions of different classifiers may prove to be optimal in currently ongoing and forthcoming photometric surveys, such as the Dark Energy Survey and the Large Synoptic Survey Telescope.

astro-ph.IM↗

arXiv2013

Narrow absorption line variability in repeat quasar observations from the Sloan Digital Sky Survey

We present the results from a time domain study of absorption lines detected in quasar spectra with repeat observations from the Sloan Digital Sky Survey Data Release 7 (SDSS DR7). Beginning with over 4500 unique time separation baselines of various absorption line species identified in the SDSS DR7 quasar spectra, we create a catalogue of 2522 quasar absorption line systems with two to eight repeat observations, representing the largest collection of unbiased and homogeneous multi-epoch absorption systems ever published. To investigate these systems for time variability of narrow absorption lines, we refine this sample based on the reliability of the system detection, the proximity of pixels with bright sky contamination to individual absorption lines, and the quality of the continuum fit. Variability measurements of this sub-sample based on the absorption line equivalent widths yield a total of 33 systems with indications of significantly variable absorption strengths on time-scales ranging from one day to several years in the rest frame of the absorption system. Of these, at least 10 are from a class known as intervening absorption systems caused by foreground galaxies along the line of sight to the background quasar. This is the first evidence of possible absorption line variability detected in intervening systems, and their short time-scale variations suggest that small-scale structures (~10-100 au) are likely to exist in their host foreground galaxies.

astro-ph.CO↗

arXiv2010

Data Mining and Machine Learning in Astronomy

We review the current state of data mining and machine learning in astronomy. 'Data Mining' can have a somewhat mixed connotation from the point of view of a researcher in this field. If used correctly, it can be a powerful approach, holding the potential to fully exploit the exponentially increasing amount of available data, promising great scientific advance. However, if misused, it can be little more than the black-box application of complex computing algorithms that may give little physical insight, and provide questionable results. Here, we give an overview of the entire data mining process, from data collection through to the interpretation of results. We cover common machine learning algorithms, such as artificial neural networks and support vector machines, applications from a broad range of astronomy, emphasizing those where data mining techniques directly resulted in improved science, and important current and future directions, including probability density functions, parallel algorithms, petascale computing, and the time domain. We conclude that, so long as one carefully selects an appropriate algorithm, and is guided by the astronomical problem at hand, data mining can be very much the powerful tool, and not the questionable black box.

astro-ph.IM↗

arXiv2010

Evolution of the Clustering of Photometrically Selected SDSS Galaxies

We measure the angular auto-correlation functions (w) of SDSS galaxies selected to have photometric redshifts 0.1 < z < 0.4 and absolute r-band magnitudes Mr < -21.2. We split these galaxies into five overlapping redshift shells of width 0.1 and measure w in each subsample in order to investigate the evolution of SDSS galaxies. We find that the bias increases substantially with redshift - much more so than one would expect for a passively evolving sample. We use halo-model analysis to determine the best-fit halo-occupation-distribution (HOD) for each subsample, and the best-fit models allow us to interpret the change in bias physically. In order to properly interpret our best-fit HODs, we convert each halo mass to its z = 0 passively evolved bias (bo), enabling a direct comparison of the best-fit HODs at different redshifts. We find that the minimum halo bo required to host a galaxy decreases as the redshift decreases, suggesting that galaxies with Mr < -21.2 are forming in halos at the low-mass end of the HODs over our redshift range. We use the best-fit HODs to determine the change in occupation number divided by the change in mass of halos with constant bo and we find a sharp peak at bo ~ 0.9 - corresponding to an average halo mass of ~ 10^12Msol/h. We thus present the following scenario: the bias of galaxies with Mr < -21.2 decreases as the Universe evolves because these galaxies form in halos of mass ~ 10^12Msol/h (independent of redshift), and the bias of these halos naturally decreases as the Universe evolves.

astro-ph.CO↗

arXiv2009

LSST Science Book, Version 2.0

A survey that can cover the sky in optical bands over wide fields to faint magnitudes with a fast cadence will enable many of the exciting science opportunities of the next decade. The Large Synoptic Survey Telescope (LSST) will have an effective aperture of 6.7 meters and an imaging camera with field of view of 9.6 deg^2, and will be devoted to a ten-year imaging survey over 20,000 deg^2 south of +15 deg. Each pointing will be imaged 2000 times with fifteen second exposures in six broad bands from 0.35 to 1.1 microns, to a total point-source depth of r~27.5. The LSST Science Book describes the basic parameters of the LSST hardware, software, and observing plans. The book discusses educational and outreach opportunities, then goes on to describe a broad range of science that LSST will revolutionize: mapping the inner and outer Solar System, stellar populations in the Milky Way and nearby galaxies, the structure of the Milky Way disk and halo and other objects in the Local Volume, transient and variable objects both at low and high redshift, and the properties of normal and active galaxies at low and high redshift. It then turns to far-field cosmological topics, exploring properties of supernovae to z~1, strong and weak lensing, the large-scale distribution of galaxies and baryon oscillations, and how these different probes may be combined to constrain cosmological models and the physics of dark energy.

astro-ph.IM↗

arXiv2009

Halo-model Analysis of the Clustering of Photometrically Selected Galaxies from SDSS

We measure the angular 2-point correlation functions of galaxies in a volume limited, photometrically selected galaxy sample from the fifth data release of the Sloan Digital Sky Survey. We split the sample both by luminosity and galaxy type and use a halo-model analysis to find halo-occupation distributions that can simultaneously model the clustering of all, early-, and late-type galaxies in a given sample. Our results for the full galaxy sample are generally consistent with previous results using the SDSS spectroscopic sample, taking the differences between the median redshifts of the photometric and spectroscopic samples into account. We find that our early- and late- type measurements cannot be fit by a model that allows early- and late-type galaxies to be well-mixed within halos. Instead, we introduce a new model that segregates early- and late-type galaxies into separate halos to the maximum allowed extent. We determine that, in all cases, it provides a good fit to our data and thus provides a new statistical description of the manner in which early- and late-type galaxies occupy halos.

astro-ph.CO↗

arXiv2009

A Cross-Correlation Analysis of Mg II Absorption Line Systems and Luminous Red Galaxies from the SDSS DR5

We analyze the cross-correlation of 2,705 unambiguously intervening Mg II (2796,2803A) quasar absorption line systems with 1,495,604 luminous red galaxies (LRGs) from the Fifth Data Release of the Sloan Digital Sky Survey within the redshift range 0.36<=z<=0.8. We confirm with high precision a previously reported weak anti-correlation of equivalent width and dark matter halo mass, measuring the average masses to be log M_h(M_[solar]h^-1)=11.29 [+0.36,-0.62] and log M_h(M_[solar]h^-1)=12.70 [+0.53,-1.16] for systems with W[2796A]>=1.4A and 0.8A<=W[2796A]<1.4A, respectively. Additionally, we investigate the significance of a number of potential sources of bias inherent in absorber-LRG cross-correlation measurements, including absorber velocity distributions and the weak lensing of background quasars, which we determine is capable of producing a 20-30% bias in angular cross-correlation measurements on scales less than 2'. We measure the Mg II - LRG cross-correlation for 719 absorption systems with v<60,000 km s^-1 in the quasar rest frame and find that these associated absorbers typically reside in dark matter haloes that are ~10-100 times more massive than those hosting unambiguously intervening Mg II absorbers. Furthermore, we find evidence for evolution of the redshift number density, dN/dz, with 2-sigma significance for the strongest (W>2.0A) absorbers in the DR5 sample. This width-dependent dN/dz evolution does not significantly affect the recovered equivalent width-halo mass anti-correlation and adds to existing evidence that the strongest Mg II absorption systems are correlated with an evolving population of field galaxies at z<0.8, while the non-evolving dN/dz of the weakest absorbers more closely resembles that of the LRG population.

astro-ph.CO↗

arXiv2009

Clustering of Low-Redshift (z <= 2.2) Quasars from the Sloan Digital Sky Survey

We present measurements of the quasar two-point correlation function, ξ_{Q}, over the redshift range z=0.3-2.2 based upon data from the SDSS. Using a homogeneous sample of 30,239 quasars with spectroscopic redshifts from the DR5 Quasar Catalogue, our study represents the largest sample used for this type of investigation to date. With this redshift range and an areal coverage of approx 4,000 deg^2, we sample over 25 h^-3 Gpc^3 (comoving) assuming the current LCDM cosmology. Over this redshift range, we find that the redshift-space correlation function, xi(s), is adequately fit by a single power-law, with s_{0}=5.95+/-0.45 h^-1 Mpc and γ_{s}=1.16+0.11-0.16 when fit over s=1-25 h^-1 Mpc. Using the projected correlation function we calculate the real-space correlation length, r_{0}=5.45+0.35-0.45 h^-1 Mpc and γ=1.90+0.04-0.03, over scales of rp=1-130 h^-1 Mpc. Dividing the sample into redshift slices, we find very little, if any, evidence for the evolution of quasar clustering, with the redshift-space correlation length staying roughly constant at s_{0} ~ 6-7 h^-1 Mpc at z<2.2 (and only increasing at redshifts greater than this). Comparing our clustering measurements to those reported for X-ray selected AGN at z=0.5-1, we find reasonable agreement in some cases but significantly lower correlation lengths in others. We find that the linear bias evolves from b~1.4 at z=0.5 to b~3 at z=2.2, with b(z=1.27)=2.06+/-0.03 for the full sample. We compare our data to analytical models and infer that quasars inhabit dark matter haloes of constant mass M ~2 x 10^12 h^-1 M_Sol from redshifts z~2.5 (the peak of quasar activity) to z~0. [ABRIDGED]

astro-ph.CO↗

arXiv2009

Eight-Dimensional Mid-Infrared/Optical Bayesian Quasar Selection

We explore the multidimensional, multiwavelength selection of quasars from mid-IR (MIR) plus optical data, specifically from Spitzer-IRAC and the Sloan Digital Sky Survey (SDSS). We apply modern statistical techniques to combined Spitzer MIR and SDSS optical data, allowing up to 8-D color selection of quasars. Using a Bayesian selection method, we catalog 5546 quasar candidates to an 8.0 um depth of 56 uJy over an area of ~24 sq. deg; ~70% of these candidates are not identified by applying the same Bayesian algorithm to 4-color SDSS optical data alone. Our selection recovers 97.7% of known type 1 quasars in this area and greatly improves the effectiveness of identifying 3.5<z<5 quasars. Even using only the two shortest wavelength IRAC bandpasses, it is possible to use our Bayesian techniques to select quasars with 97% completeness and as little as 10% contamination. This sample has a photometric redshift accuracy of 93.6% (Delta Z +/-0.3), remaining roughly constant when the two reddest MIR bands are excluded. While our methods are designed to find type 1 (unobscured) quasars, as many as 1200 of the objects are type 2 (obscured) quasar candidates. Coupling deep optical imaging data with deep mid-IR data could enable selection of quasars in significant numbers past the peak of the quasar luminosity function (QLF) to at least z~4. Such a sample would constrain the shape of the QLF and enable quasar clustering studies over the largest range of redshift and luminosity to date, yielding significant gains in our understanding of quasars and the evolution of galaxies.

astro-ph↗

arXiv2008

Quasar Clustering from SDSS DR5: Dependences on Physical Properties

Using a homogenous sample of 38,208 quasars with a sky coverage of $4000 {\rm deg^2}$ drawn from the SDSS Data Release Five quasar catalog, we study the dependence of quasar clustering on luminosity, virial black hole mass, quasar color, and radio loudness. At $z<2.5$, quasar clustering depends weakly on luminosity and virial black hole mass, with typical uncertainty levels $\sim 10%$ for the measured correlation lengths. These weak dependences are consistent with models in which substantial scatter between quasar luminosity, virial black hole mass and the host dark matter halo mass has diluted any clustering difference, where halo mass is assumed to be the relevant quantity that best correlates with clustering strength. However, the most luminous and most massive quasars are more strongly clustered (at the $\sim 2σ$ level) than the remainder of the sample, which we attribute to the rapid increase of the bias factor at the high-mass end of host halos. We do not observe a strong dependence of clustering strength on quasar colors within our sample. On the other hand, radio-loud quasars are more strongly clustered than are radio-quiet quasars matched in redshift and optical luminosity (or virial black hole mass), consistent with local observations of radio galaxies and radio-loud type 2 AGN. Thus radio-loud quasars reside in more massive and denser environments in the biased halo clustering picture. Using the Sheth et al.(2001) formula for the linear halo bias, the estimated host halo mass for radio-loud quasars is $\sim 10^{13} h^{-1}M_\odot$, compared to $\sim 2\times 10^{12} h^{-1}M_\odot$ for radio-quiet quasar hosts at $z\sim 1.5$.

astro-ph↗

arXiv2008

The 2dF-SDSS LRG and QSO Survey: The spectroscopic QSO catalogue

We present the final spectroscopic QSO catalogue from the 2dF-SDSS LRG and QSO (2SLAQ) Survey. This is a deep, 18<g<21.85 (extinction corrected), sample aimed at probing in detail the faint end of the broad line AGN luminosity distribution at z<2.6. The candidate QSOs were selected from SDSS photometry and observed spectroscopically with the 2dF spectrograph on the Anglo-Australian Telescope. This sample covers an area of 191.9 deg^2 and contains new spectra of 16326 objects, of which 8764 are QSOs, and 7623 are newly discovered (the remainder were previously identified by the 2QZ and SDSS surveys). The full QSO sample (including objects previously observed in the SDSS and 2QZ surveys) contains 12702 QSOs. The new 2SLAQ spectroscopic data set also contains 2343 Galactic stars, including 362 white dwarfs, and 2924 narrow emission line galaxies with a median redshift of z=0.22. We present detailed completeness estimates for the survey, based on modelling of QSO colours, including host galaxy contributions. This calculation shows that at g~21.85 QSO colours are significantly affected by the presence of a host galaxy up to redshift z~1 in the SDSS ugriz bands. In particular we see a significant reddening of the objects in g-i towards fainter g-band magnitudes. This reddening is consistent with the QSO host galaxies being dominated by a stellar population of age at least 2-3 Gyr. The full catalogue, including completeness estimates, is available on-line at http://www.2slaq.info/

astro-ph↗