arXiv subjects

M. Treyer

Publications and source records attributed to M. Treyer.

14 recordsLinked to original sources

arXiv2025

The Domain Adaptation problem in photometric redshift estimation: a solution applied to the HSC Survey

The multi-band HSC-CLAUDS survey comprises several sky regions with varying observing conditions, only one of which, the COSMOS Ultra Deep Field (UDF), offers extensive redshift coverage. We aim to exploit a complete sample of labeled galaxies from the COSMOS UDF at i<25 (z<~5) to train a convolutional neural network (CNN) and infer more accurate photometric redshifts in the other regions than those currently available from SED-fitting methods. To address the severe domain mismatch problem we observed when applying the trained CNN to regions other than the COSMOS UDF, we developed an unsupervised adversarial domain adaptation network that we grafted onto the CNN. The method is validated by three tests: the predicted redshifts are compared to the spectroscopic redshifts that are available for limited samples of mostly bright galaxies; the predicted redshift distributions of the entire galaxy population of a given field in several intervals of magnitude are compared to those of the COSMOS UDF, assumed to be representative; the redshifts predicted for a sample of galaxies selected by narrow-band filter observations sensitive to [OII] emitters at z~1.47 are compared to those of confirmed [OII] emission line galaxies. The results show successful domain adaptation: the network is able to transfer its redshift classification capability learnt from the COSMOS UDF to other regions of HSC-CLAUDS. Accuracy varies depending on magnitude and redshift, following that of the labels we used, but far exceeds that of currently available photometric redshifts. The catalogs of CNN redshifts we inferred for the XMM, DEEP2 and ELAIS fields and for the remaining COSMOS region (~4 million sources in total at i<25) are made public.

astro-ph.CO↗

arXiv2023

CNN photometric redshifts in the SDSS at $r\leq 20$

We release photometric redshifts, reaching $\sim$0.7, for $\sim$14M galaxies at $r\leq 20$ in the 11,500 deg$^2$ of the SDSS north and south galactic caps. These estimates were inferred from a convolution neural network (CNN) trained on $ugriz$ stamp images of galaxies labelled with a spectroscopic redshift from the SDSS, GAMA and BOSS surveys. Representative training sets of $\sim$370k galaxies were constructed from the much larger combined spectroscopic data to limit biases, particularly those arising from the over-representation of Luminous Red Galaxies. The CNN outputs a redshift classification that offers all the benefits of a well-behaved PDF, with a width efficiently signaling unreliable estimates due to poor photometry or stellar sources. The dispersion, mean bias and rate of catastrophic failures of the median point estimate are of order $σ_{\rm MAD}=0.014$, <$Δz_{\rm norm}$>$=0.0015$, $η(|Δz_{\rm norm}|>0.05)=4\%$ on a representative test sample at $r<19.8$, out-performing currently published estimates. The distributions in narrow intervals of magnitudes of the redshifts inferred for the photometric sample are in good agreement with the results of tomographic analyses. The inferred redshifts also match the photometric redshifts of the redMaPPer galaxy clusters for the probable cluster members. The CNN input and output are available at: https://deepdip.iap.fr/treyer+2023.

astro-ph.CO↗

arXiv2023

Multimodality for improved CNN photometric redshifts

Photometric redshift estimation plays a crucial role in modern cosmological surveys for studying the universe's large-scale structures and the evolution of galaxies. Deep learning has emerged as a powerful method to produce accurate photometric redshift estimates from multi-band images of galaxies. Here, we introduce a multimodal approach consisting of the parallel processing of several subsets of image bands prior, the outputs of which are then merged for further processing through a convolutional neural network (CNN). We evaluate the performance of our method using three surveys: the Sloan Digital Sky Survey (SDSS), The Canada-France-Hawaii Telescope Legacy Survey (CFHTLS) and Hyper Suprime-Cam (HSC). By improving the model's ability to capture information embedded in the correlation between different bands, our technique surpasses the state-of-the-art photometric redshift precision. We find that the positive gain does not depend on the specific architecture of the CNN and that it increases with the number of photometric filters available.

astro-ph.CO↗

arXiv2022

Photometric Redshift Estimation with Convolutional Neural Networks and Galaxy Images: A Case Study of Resolving Biases in Data-Driven Methods

Deep Learning models have been increasingly exploited in astrophysical studies, yet such data-driven algorithms are prone to producing biased outputs detrimental for subsequent analyses. In this work, we investigate two major forms of biases, i.e., class-dependent residuals and mode collapse, in a case study of estimating photometric redshifts as a classification problem using Convolutional Neural Networks (CNNs) and galaxy images with spectroscopic redshifts. We focus on point estimates and propose a set of consecutive steps for resolving the two biases based on CNN models, involving representation learning with multi-channel outputs, balancing the training data and leveraging soft labels. The residuals can be viewed as a function of spectroscopic redshifts or photometric redshifts, and the biases with respect to these two definitions are incompatible and should be treated in a split way. We suggest that resolving biases in the spectroscopic space is a prerequisite for resolving biases in the photometric space. Experiments show that our methods possess a better capability in controlling biases compared to benchmark methods, and exhibit robustness under varying implementing and training conditions provided with high-quality data. Our methods have promises for future cosmological surveys that require a good constraint of biases, and may be applied to regression problems and other studies that make use of data-driven models. Nonetheless, the bias-variance trade-off and the demand on sufficient statistics suggest the need for developing better methodologies and optimizing data usage strategies.

astro-ph.IM↗

arXiv2020

PhotoWeb redshift: boosting photometric redshift accuracy with large spectroscopic surveys

Improving distance measurements in large imaging surveys is a major challenge to better reveal the distribution of galaxies on a large scale and to link galaxy properties with their environments. Photometric redshifts can be efficiently combined with the cosmic web (CW) extracted from overlapping spectroscopic surveys to improve their accuracy. We apply a similar method using a new generation of photometric redshifts based on a convolution neural network (CNN). The CNN is trained on the SDSS images with the main galaxy sample (SDSS-MGS, $r \leq 17.8$) and the GAMA spectroscopic redshifts up tor $\sim 19.8$. The mapping of the CW is obtained with 680,000 spectroscopic redshifts from the MGS and BOSS surveys. The redshift probability distribution functions (PDF), which are well calibrated (unbiased and narrow, $\leq 120$ Mpc), intercept a few CW structure along the line of sight. Combining these PDFs with the density field distribution provides new photometric redshifts, $z_{web}$, whose accuracy is improved by a factor of two (i.e.,$σ \sim 0.004(1+z)$) for galaxies with $r \leq 17.8$. For half of them, the distance accuracy is better than 10 cMpc. The narrower the original PDF, the larger the boost in accuracy. No gain is observed for original PDFs wider than 0.03. The final $z_{web}$ PDFs also appear well calibrated. The method performs slightly better for passive galaxies than star-forming ones, and for galaxies in massive groups since these populations better trace the underlying large-scale structure. Reducing the spectroscopic sampling by a factor of 8 still improves the photometric redshift accuracy by 25%. Extending the method to galaxies fainter than the MGS limit still improves the redshift estimates for 70% of the galaxies, with a gain in accuracy of 20% at low $z$ where the resolution of the CW is the highest.

astro-ph.GA↗

arXiv2018

Galaxies flowing in the oriented saddle frame of the cosmic web

The strikingly anisotropic large-scale distribution of matter made of an extended network of voids delimited by sheets, themselves segmented by filaments, within which matter flows towards compact nodes where they intersect, imprints its geometry on the dynamics of cosmic flows, ultimately shaping the distribution of galaxies and the redshift evolution of their properties. The (filament-type) saddle points of this cosmic web provide a local frame in which to quantify the induced physical and morphological evolution of galaxies on large scales. The properties of virtual galaxies within the Horizon-AGN simulation are stacked in such a frame. The iso-contours of the galactic number density, mass, specific star formation rate (sSFR), kinematics and age are clearly aligned with the filament axis with steep gradients perpendicular to the filaments. A comparison to a simulation without feedback from active galactic nuclei (AGN) illustrates its impact on quenching star formation of centrals away from the saddles. The redshift evolution of the properties of galaxies and their age distribution are consistent with the geometry of the bulk flow within that frame. They compare well with expectations from constrained Gaussian random fields and the scaling with the mass of non-linearity, modulo the redshift dependent impact of feedback processes. Physical properties such as sSFR and kinematics seem not to depend only on mean halo mass and density: the residuals trace the geometry of the saddle, which could point to other environment-sensitive physical processes, such as spin advection, and AGN feedback at high mass.

astro-ph.GA↗

arXiv2018

Group quenching and galactic conformity at low redshift

We quantify the quenching impact of the group environment using the spectroscopic survey Galaxy and Mass Assembly (GAMA) to z=0.2. The fraction of red (quiescent) galaxies, whether in groups or isolated, increases with both stellar mass and large-scale (5 Mpc) density. At fixed stellar mass, the red fraction is on average higher for satellites of red centrals than of blue (star-forming) centrals, a galactic conformity effect that increases with density. Most of the signal originates from groups that have the highest stellar mass, reside in the densest environments, and have massive, red only centrals. Assuming a color-dependent halo-to-stellar-mass ratio, whereby red central galaxies inhabit significantly more massive halos than blue ones of the same stellar mass, two regimes emerge more distinctly: at log(Mhalo/Msol) < 13, central quenching is still ongoing, conformity is no longer existent, and satellites and group centrals exhibit the same quenching excess over field galaxies at all mass and density, in agreement with the concept of "group quenching"; at log(Mhalo/Msol) > 13, a cutoff that sets apart massive (log(M*/Msol) > 11), fully quenched group centrals, conformity is meaningless, and satellites undergo significantly more quenching than their counterparts in smaller halos. The latter effect strongly increases with density, giving rise to the density-dependent conformity signal when both regimes are mixed. The star-formation of blue satellites in massive halos is also suppressed compared to blue field galaxies, while blue group centrals and the majority of blue satellites, which reside in low mass halos, show no deviation from the color-stellar mass relation of blue field galaxies.

astro-ph.GA↗

arXiv2017

Galaxy evolution in the metric of the Cosmic Web

The role of the cosmic web in shaping galaxy properties is investigated in the GAMA spectroscopic survey in the redshift range $0.03 \leq z \leq 0.25$. The stellar mass, $u - r$ dust corrected colour and specific star formation rate (sSFR) of galaxies are analysed as a function of their distances to the 3D cosmic web features, such as nodes, filaments and walls, as reconstructed by DisPerSE. Significant mass and type/colour gradients are found for the whole population, with more massive and/or passive galaxies being located closer to the filament and wall than their less massive and/or star-forming counterparts. Mass segregation persists among the star-forming population alone. The red fraction of galaxies increases when closing in on nodes, and on filaments regardless of the distance to nodes. Similarly, the star-forming population reddens (or lowers its sSFR) at fixed mass when closing in on filament, implying that some quenching takes place. Comparable trends are also found in the state-of-the-art hydrodynamical simulation Horizon-AGN. These results suggest that on top of stellar mass and large-scale density, the traceless component of the tides from the anisotropic large-scale environment also shapes galactic properties. An extension of excursion theory accounting for filamentary tides provides a qualitative explanation in terms of anisotropic assembly bias: at a given mass, the accretion rate varies with the orientation and distance to filaments. It also explains the absence of type/colour gradients in the data on smaller, non-linear scales.

astro-ph.GA↗

arXiv2016

The VIMOS Public Extragalactic Redshift Survey (VIPERS): Galaxy segregation inside filaments at $z \simeq 0.7$

We present the first quantitative detection of large-scale filamentary structure at $z \simeq 0.7$ in the large cosmological volume probed by the VIMOS Public Extragalactic Redshift Survey (VIPERS). We use simulations to show the capability of VIPERS to recover robust topological features in the galaxy distribution, in particular the filamentary network. We then investigate how galaxies with different stellar masses and stellar activities are distributed around the filaments and find a significant segregation, with the most massive or quiescent galaxies being closer to the filament axis than less massive or active galaxies. The signal persists even after down-weighting the contribution of peak regions. Our results suggest that massive and quiescent galaxies assemble their stellar mass through successive mergers during their migration along filaments towards the nodes of the cosmic web. On the other hand, low-mass star-forming galaxies prefer the outer edge of filaments, a vorticity rich region dominated by smooth accretion, as predicted by the recent spin alignment theory. This emphasizes the role of large scale cosmic flows in shaping galaxy properties.

astro-ph.GA↗

arXiv2016

The VIPERS Multi-Lambda Survey. I. UV and NIR Observations, multi-color catalogues and photometric redshifts

We present observations collected in the CFHTLS-VIPERS region in the ultraviolet (UV) with the GALEX satellite (far and near UV channels) and the near infrared with the CFHT/WIRCam camera ($K_s$-band) over an area of 22 and 27 deg$^2$, respectively. The depth of the photometry was optimized to measure the physical properties (e.g., SFR, stellar masses) of all the galaxies in the VIPERS spectroscopic survey. The large volume explored by VIPERS will enable a unique investigation of the relationship between the galaxy properties and their environment (density field and cosmic web) at high redshift (0.5 < z < 1.2). In this paper, we present the observations, the data reductions and the build-up of the multi-color catalogues. The CFHTLS-T0007 (gri-χ^2) images are used as reference to detect and measure the $K_s$-band photometry, while the T0007 u-selected sources are used as priors to perform the GALEX photometry based on a dedicated software (EMphot). Our final sample reaches $NUV_{AB}$~25 (at 5σ) and $K_{AB}$~22 (at 3σ). The large spectroscopic sample (~51,000 spectroscopic redshifts) allows us to highlight the robustness of our star/galaxy separation, and the reliability of our photometric redshifts with a typical accuracy $σ_z \le$ 0.04 and a catastrophic failure rate η < 2% down to i~23. We present various tests on the $K_s$ band completeness and photometric redshift accuracy by comparing with existing, overlapping deep photometric catalogues. Finally, we discuss the BzK sample of passive and active galaxies at high redshift and the evolution of galaxy morphology in the (NUV-r) vs (r-K_s) diagram at low redshift (z < 0.25) thanks to the high image quality of the CFHTLS. The images, catalogues and photometric redshifts for 1.5 million sources (down to $NUV \le$ 25 or $K_s \le$ 22) are released and available at this URL: http://cesam.lam.fr/vipers-mls/

astro-ph.GA↗

arXiv2004

The GALEX-VVDS Measurement of the Evolution of the 1500A Luminosity Function

We present the first measurement of the galaxy luminosity function at 1500A between 0.2<z<1.2 based on GALEX-VVDS observations (1000 spectroscopic redshifts for galaxies with NUV<24.5) and at higher z using existing datasets. Our main results are summarized as follows : (i) luminosity evolution is observed with Delta(Mstar)=-2.0 mag between z=0 and z=1 and Delta(Mstar)=-1.0mag between z=1 and z=3. This confirms that the star formation activity was significantly higher in the past. (ii) the LF slopes vary between -1.2< alpha <-1.65, with a marginally significant hint of increase at higher z. (iii) we split the sample in three restframe (B-I) intervals providing an approximate spectral type classification: Sb-Sd, Sd-Irr and unobscured starbursts. We find that the bluest class evolves less strongly in luminosity than the two other classes. On the other hand their number density increases sharply with z (15% in the local universe to 55% at z=1) while that of the reddest classes decreases.

astro-ph↗

arXiv2004

The GALEX-VVDS Measurement of the Evolution of the Far-Ultraviolet Luminosity Density and the Cosmic Star Formation Rate

In a companion paper (Arnouts et al. 2004) we presented new measurements of the galaxy luminosity function at 1500 Angstroms out to z~1 using GALEX-VVDS observations (1039 galaxies with NUV<24.5 and z>0.2) and at higher z using existing data sets. In this paper we use the same sample to study evolution of the FUV luminosity density. We detect evolution consistent with a (1+z)^{2.5+/-0.7} rise to z~1 and (1+z)^{0.5+/-0.4} for z>1. The luminosity density from the most UV-luminous galaxies (UVLG) is undergoing dramatic evolution (x30) between 0 25%) of the total FUV luminosity density at z<1. We measure dust attenuation and star formation rates of our sample galaxies and determine the star formation rate density as a function of redshift, both uncorrected and corrected for dust. We find good agreement with other measures of the SFR density in the rest ultraviolet and Halpha given the still significant uncertainties in the attenuation correction.

astro-ph↗

arXiv2004

The Ultraviolet Galaxy Luminosity Function from GALEX data: Color Dependent Evolution at Low Redshift

We present measurements of the FUV (1530A) and NUV (2310A) galaxy luminosity functions (LF) at low redshift (z<0.2) from GALEX observations matched to the 2dF Galaxy Redshift Survey. We split our FUV and NUV samples into two UV-bj color bins and two redshift bins. As observed at optical wavelengths, the local LF of the bluest galaxies tend to have steeper faint end slopes and fainter characteristic magnitudes M* than the reddest subsamples. We find evidence for color dependent evolution at very low redshift in both bands, with bright blue galaxies becoming dominant in the highest redshift bin. The evolution of the total LF is consistent with an 0.3 magnitude brightening between z=0 and 0.13, in agreement with the first analysis of deeper GALEX fields probing adjacent and higher redshifts.

astro-ph↗

arXiv2000

The 2-10 keV XRB dipole and its cosmological implications

The hard X-ray (>2 keV) emission of the local and distant Universe as observed with the HEAO1-A2 experiment is reconsidered in the context of large scale cosmic structure. Using all-sky X-ray samples of AGN and galaxy clusters we remove the dominant local X-ray flux from within a redshift of ~ 0.02. We evaluate the dipolar and higher order harmonic structure in 4 X-ray colours. The estimated dipole anisotropy of the unresolved flux appears to be consistent with a combination of the Compton-Getting effect due to the Local Group motion (dipole amplitude Delta = 0.0042) and remaining large scale structure (0.0023 <~ Delta <~ 0.0085), in good agreement with the expectations of Cold Dark Matter models. The observed anisotropy does however also suggest a non-negligible Galactic contribution which is more complex than current, simple models of >2 keV Galactic X-ray emission. Comparison of the soft and hard colour maps with a harmonic analysis of the 1.5 keV ROSAT all-sky data qualitatively suggests that at least a third of the faint, unresolved ~ 18 deg scale structure in the HEAO1-A2 data may be Galactic in origin. However, the effect on measured flux dipoles is small (<~3%). We derive an expression for dipole anisotropy and acceleration and demonstrate how the dipole anisotropy of the distant X-ray frame can constrain the amplitude of bulk motions of the universe. From observed bulk motions over a local ~ 50 Mpc/h radius volume we determine 0.14 <~ Omega^0.6/b_x(0) <~ 0.59.

astro-ph↗