arXiv subjects

F. Tarsitano

Publications and source records attributed to F. Tarsitano.

24 records · Page 2Linked to original sources

arXiv2021

Image feature extraction and galaxy classification: a novel and efficient approach with automated machine learning

In this work we explore the possibility of applying machine learning methods designed for one-dimensional problems to the task of galaxy image classification. The algorithms used for image classification typically rely on multiple costly steps, such as the Point Spread Function (PSF) deconvolution and the training and application of complex Convolutional Neural Networks (CNN) of thousands or even millions of parameters. In our approach, we extract features from the galaxy images by analysing the elliptical isophotes in their light distribution and collect the information in a sequence. The sequences obtained with this method present definite features allowing a direct distinction between galaxy types, as opposed to smooth Sérsic profiles. Then, we train and classify the sequences with machine learning algorithms, designed through the platform Modulos AutoML, and study how they optimize the classification task. As a demonstration of this method, we use the second public release of the Dark Energy Survey (DES DR2). We show that by applying it to this sample we are able to successfully distinguish between early-type and late-type galaxies, for images with signal-to-noise ratio greater then 300. This yields an accuracy of $86\%$ for the early-type galaxies and $93\%$ for the late-type galaxies, which is on par with most contemporary automated image classification approaches. Our novel method allows for galaxy images to be accurately classified and is faster than other approaches. Data dimensionality reduction also implies a significant lowering in computational cost. In the perspective of future data sets obtained with e.g. Euclid and the Vera Rubin Observatory (VRO), this work represents a path towards using a well-tested and widely used platform from industry in efficiently tackling galaxy classification problems at the peta-byte scale.

astro-ph.GA↗

arXiv2021

Pushing automated morphological classifications to their limits with the Dark Energy Survey

We present morphological classifications of $\sim$27 million galaxies from the Dark Energy Survey (DES) Data Release 1 (DR1) using a supervised deep learning algorithm. The classification scheme separates: (a) early-type galaxies (ETGs) from late-types (LTGs); and (b) face-on galaxies from edge-on. Our Convolutional Neural Networks (CNNs) are trained on a small subset of DES objects with previously known classifications. These typically have $\mathrm{m}_r \lesssim 17.7~\mathrm{mag}$; we model fainter objects to $\mathrm{m}_r < 21.5$ mag by simulating what the brighter objects with well determined classifications would look like if they were at higher redshifts. The CNNs reach 97\% accuracy to $\mathrm{m}_r<21.5$ on their training sets, suggesting that they are able to recover features more accurately than the human eye. We then used the trained CNNs to classify the vast majority of the other DES images. The final catalog comprises five independent CNN predictions for each classification scheme, helping to determine if the CNN predictions are robust or not. We obtain secure classifications for $\sim$ 87\% and 73\% of the catalog for the ETG vs. LTG and edge-on vs. face-on models, respectively. Combining the two classifications (a) and (b) helps to increase the purity of the ETG sample and to identify edge-on lenticular galaxies (as ETGs with high ellipticity). Where a comparison is possible, our classifications correlate very well with Sérsic index (\textit{n}), ellipticity ($ε$) and spectral type, even for the fainter galaxies. This is the largest multi-band catalog of automated galaxy morphologies to date.

astro-ph.GA↗

arXiv2020

Predicting Cosmological Observables with PyCosmo

Current and upcoming cosmological experiments open a new era of precision cosmology, thus demanding accurate theoretical predictions for cosmological observables. Because of the complexity of the codes delivering such predictions, reaching a high level of numerical accuracy is challenging. Among the codes already fulfilling this task, $\textsf{PyCosmo}$ is a Python based framework providing solutions to the Einstein-Boltzmann equations and accurate predictions for cosmological observables. In this work, we first describe how the observables are implemented. Then, we check the accuracy of the theoretical predictions for background quantities, power spectra and Limber and beyond-Limber angular power spectra by comparison with other codes: the Core Cosmology Library ($\texttt{CCL}$), $\texttt{CLASS}$, $\texttt{HMCode}$ and $\texttt{iCosmo}$. In our analysis we quantify the agreement of $\textsf{PyCosmo}$ with the other codes, for a range of cosmological models, monitored through a series of $\textit{unit tests}$. $\textsf{PyCosmo}$, conceived as a multi purpose cosmology calculation tool in $\texttt{Python}$, is designed to be interactive and user friendly. A current version of the code (without the Boltzmann Solver) is publicly available and can be used interactively on the platform $\textsf{PyCosmo Hub}$, all accessible from this link: https://cosmology.ethz.ch/research/software-lab/PyCosmo.html . On the hub the users can perform their own computations using $\texttt{Jupyter Notebooks}$ without the need of installing any software, access to the results presented in this work and benefit from tutorial notebooks illustrating the usage of the code. The link above also redirects to the code release and documentation.

astro-ph.CO↗

arXiv2019

Monte Carlo Control Loops for cosmic shear cosmology with DES Year 1

Weak lensing by large-scale structure is a powerful probe of cosmology and of the dark universe. This cosmic shear technique relies on the accurate measurement of the shapes and redshifts of background galaxies and requires precise control of systematic errors. The Monte Carlo Control Loops (MCCL) is a forward modelling method designed to tackle this problem. It relies on the Ultra Fast Image Generator (UFig) to produce simulated images tuned to match the target data statistically, followed by calibrations and tolerance loops. We present the first end-to-end application of this method, on the Dark Energy Survey (DES) Year 1 wide field imaging data. We simultaneously measure the shear power spectrum $C_{\ell}$ and the redshift distribution $n(z)$ of the background galaxy sample. The method includes maps of the systematic sources, Point Spread Function (PSF), an Approximate Bayesian Computation (ABC) inference of the simulation model parameters, a shear calibration scheme, and the fast estimation of the covariance matrix. We find a close statistical agreement between the simulations and the DES Y1 data using an array of diagnostics. In a non-tomographic setting, we derive a set of $C_\ell$ and $n(z)$ curves that encode the cosmic shear measurement, as well as the systematic uncertainty. Following a blinding scheme, we measure the combination of $Ω_m$, $σ_8$, and intrinsic alignment amplitude $A_{\rm{IA}}$, defined as $S_8D_{\rm{IA}} = σ_8(Ω_m/0.3)^{0.5}D_{\rm{IA}}$, where $D_{\rm{IA}}=1-0.11(A_{\rm{IA}}-1)$. We find $S_8D_{\rm{IA}}=0.895^{+0.054}_{-0.039}$, where systematics are at the level of roughly 60\% of the statistical errors. We discuss these results in the context of earlier cosmic shear analyses of the DES Y1 data. Our findings indicate that this method and its fast runtime offer good prospects for cosmic shear measurements with future wide-field surveys.

astro-ph.CO↗

arXiv2018

A catalogue of structural and morphological measurements for DES Y1

We present a structural and morphological catalogue for 45 million objects selected from the first year of data from the Dark Energy Survey (DES). Single Sersic fits and non-parametric measurements are produced for g, r and i filters. The parameters from the best-fitting Sersic model (total magnitude, half-light radius, Sersic index, axis ratio and position angle) are measured with Galfit; the non-parametric coefficients (concentration, asymmetry, clumpiness, Gini, M20) are provided using the Zurich Estimator of Structural Types (ZEST+). To study the statistical uncertainties, we consider a sample of state-of-the-art image simulations with a realistic distribution in the input parameter space and then process and analyse them as we do with real data: this enables us to quantify the observational biases due to PSF blurring and magnitude effects and correct the measurements as a function of magnitude, galaxy size, Sersic index (concentration for the analysis of the non-parametric measurements) and ellipticity. We present the largest structural catalogue to date: we find that accurate and complete measurements for all the structural parameters are typically obtained for galaxies with SExtractor MAG AUTO I < 21. Indeed, the parameters in the filters i and r can be overall well recovered up to MAG AUTO < 21.5, corresponding to a fitting completeness of ~90% below this threshold, for a total of 25 million galaxies. The combination of parametric and non-parametric structural measurements makes this catalogue an important instrument to explore and understand how galaxies form and evolve. The catalogue described in this paper will be publicly released alongside the Dark Energy Survey collaboration Y1 cosmology data products at the following URL: https://des.ncsa.illinois.edu/releases/y1a1/gold/morphology.

astro-ph.GA↗

arXiv2017

Evidence for Dynamically Driven Formation of the GW170817 Neutron Star Binary in NGC 4993

We present a study of NGC 4993, the host galaxy of the GW170817 gravitational wave event, the GRB170817A short gamma-ray burst (sGRB) and the AT2017gfo kilonova. We use Dark Energy Camera imaging, AAT spectra and publicly available data, relating our findings to binary neutron star (BNS) formation scenarios and merger delay timescales. NGC4993 is a nearby (40 Mpc) early-type galaxy, with $i$-band Sérsic index $n=4.0$ and low asymmetry ($A=0.04\pm 0.01$). These properties are unusual for sGRB hosts. However, NGC4993 presents shell-like structures and dust lanes indicative of a recent galaxy merger, with the optical transient located close to a shell. We constrain the star formation history (SFH) of the galaxy assuming that the galaxy merger produced a star formation burst, but find little to no on-going star formation in either spatially-resolved broadband SED or spectral fitting. We use the best-fit SFH to estimate the BNS merger rate in this type of galaxy, as $R_{NSM}^{gal}= 5.7^{+0.57}_{-3.3} \times 10^{-6} {\rm yr}^{-1}$. If star formation is the only considered BNS formation scenario, the expected number of BNS mergers from early-type galaxies detectable with LIGO during its first two observing seasons is $0.038^{+0.004}_{-0.022}$, as opposed to $\sim 0.5$ from all galaxy types. Hypothesizing that the binary system formed due to dynamical interactions during the galaxy merger, the subsequent time elapsed can constrain the delay time of the BNS coalescence. By using velocity dispersion estimates and the position of the shells, we find that the galaxy merger occurred $t_{\rm mer}\lesssim 200~{\rm Myr}$ prior to the BNS coalescence.

astro-ph.HE↗