Search

This paper presents a light-weight and accurate deep neural model for audiovisual emotion recognition. To design this model, the authors followed a philosophy of simplicity, drastically limiting the number of parameters to learn from the target datasets, always choosing the simplest earning methods: i) transfer learning and low-dimensional space embedding allows to reduce the dimensionality of the representations. ii) The isual temporal information is handled by a simple score-per-frame selection process, averaged across time. iii) A simple frame selection echanism is also proposed to weight the images of a sequence. iv) The fusion of the different modalities is performed at prediction level (late usion). We also highlight the inherent challenges of the AFEW dataset and the difficulty of model selection with as few as 383 validation equences. The proposed real-time emotion classifier achieved a state-of-the-art accuracy of 60.64 % on the test set of AFEW, and ranked 4th at he Emotion in the Wild 2018 challenge.

Numerous models describing the human emotional states have been built by the psychology community. Alongside, Deep Neural Networks (DNN) are reaching excellent performances and are becoming interesting features extraction tools in many computer vision tasks.Inspired by works from the psychology community, we first study the link between the compact two-dimensional representation of the emotion known as arousal-valence, and discrete emotion classes (e.g. anger, happiness, sadness, etc.) used in the computer vision community. It enables to assess the benefits -- in terms of discrete emotion inference -- of adding an extra dimension to arousal-valence (usually named dominance). Building on these observations, we propose CAKE, a 3-dimensional representation of emotion learned in a multi-domain fashion, achieving accurate emotion recognition on several public datasets. Moreover, we visualize how emotions boundaries are organized inside DNN representations and show that DNNs are implicitly learning arousal-valence-like descriptions of emotions. Finally, we use the CAKE representation to compare the quality of the annotations of different public datasets.

We explore the impact of an update to the typical approximation for the shape noise term in the analytic covariance matrix for cosmic shear experiments that assumes the absence of survey boundary and mask effects. We present an exact expression for the number of galaxy pairs in this term based on the the survey mask, which leads to more than a factor of three increase in the shape noise on the largest measured scales for the Kilo-Degree Survey (KIDS-450) real-space cosmic shear data. We compare the result of this analytic expression to several alternative methods for measuring the shape noise from the data and find excellent agreement. This update to the covariance resolves any internal model tension evidenced by the previously large cosmological best-fit $\chi^2$ for the KiDS-450 cosmic shear data. The best-fit $\chi^2$ is reduced from 161 to 121 for 118 degrees of freedom. We also apply a correction to how the multiplicative shear calibration uncertainty is included in the covariance. This change, along with a previously known update to the reported effective angular values of the data vector, jointly shift the inferred amplitude of the correlation function to higher values. We find that this improves agreement of the KiDS-450 cosmic shear results with Dark Energy Survey Year 1 and Planck results., Comment: 7 pages, 2 figures; matches accepted version

We present calibrations of the redshift distributions of redMaGiC galaxies in the Dark Energy Survey Year 1 (DES Y1) and Sloan Digital Sky Survey (SDSS) DR8 data. These results determine the priors of the redshift distribution of redMaGiC galaxies, which were used for galaxy clustering measurements and as lenses for galaxy-galaxy lensing measurements in DES Y1 cosmological analyses. We empirically determine the bias in redMaGiC photometric redshift estimates using angular cross-correlations with Baryon Oscillation Spectroscopic Survey (BOSS) galaxies. For DES, we calibrate a single parameter redshift bias in three photometric redshift bins: $z \in[0.15,0.3]$, [0.3,0.45], and [0.45,0.6]. Our best fit results in each bin give photometric redshift biases of $|\Delta z|<0.01$. To further test the redMaGiC algorithm, we apply our calibration procedure to SDSS redMaGiC galaxies, where the statistical precision of the cross-correlation measurement is much higher due to a greater overlap with BOSS galaxies. For SDSS, we also find best fit results of $|\Delta z|<0.01$. We compare our results to other analyses of redMaGiC photometric redshifts., Comment: 18 pages, 12 figures. Minor changes to text to match version accepted by MNRAS

We present the calibration of the Dark Energy Survey Year 1 (DES Y1) weak lensing source galaxy redshift distributions from clustering measurements. By cross-correlating the positions of source galaxies with luminous red galaxies selected by the redMaGiC algorithm we measure the redshift distributions of the source galaxies as placed into different tomographic bins. These measurements constrain any such shifts to an accuracy of $\sim0.02$ and can be computed even when the clustering measurements do not span the full redshift range. The highest-redshift source bin is not constrained by the clustering measurements because of the minimal redshift overlap with the redMaGiC galaxies. We compare our constraints with those obtained from $\texttt{COSMOS}$ 30-band photometry and find that our two very different methods produce consistent constraints., Comment: 13 pages, 7 figures. Submitting to MNRAS. Feedback welcome!

This paper addresses the question of emotion classification. The task consists in predicting emotion labels (taken among a set of possible labels) best describing the emotions contained in short video clips. Building on a standard framework -- lying in describing videos by audio and visual features used by a supervised classifier to infer the labels -- this paper investigates several novel directions. First of all, improved face descriptors based on 2D and 3D Convo-lutional Neural Networks are proposed. Second, the paper explores several fusion methods, temporal and multimodal, including a novel hierarchical method combining features and scores. In addition, we carefully reviewed the different stages of the pipeline and designed a CNN architecture adapted to the task; this is important as the size of the training set is small compared to the difficulty of the problem, making generalization difficult. The so-obtained model ranked 4th at the 2017 Emotion in the Wild challenge with the accuracy of 58.8 %.

We use numerical simulations to characterize the performance of a clustering-based method to calibrate photometric redshift biases. In particular, we cross-correlate the weak lensing (WL) source galaxies from the Dark Energy Survey Year 1 (DES Y1) sample with redMaGiC galaxies (luminous red galaxies with secure photometric redshifts) to estimate the redshift distribution of the former sample. The recovered redshift distributions are used to calibrate the photometric redshift bias of standard photo-$z$ methods applied to the same source galaxy sample. We apply the method to three photo-$z$ codes run in our simulated data: Bayesian Photometric Redshift (BPZ), Directional Neighborhood Fitting (DNF), and Random Forest-based photo-$z$ (RF). We characterize the systematic uncertainties of our calibration procedure, and find that these systematic uncertainties dominate our error budget. The dominant systematics are due to our assumption of unevolving bias and clustering across each redshift bin, and to differences between the shapes of the redshift distributions derived by clustering vs photo-$z$'s. The systematic uncertainty in the mean redshift bias of the source galaxy sample is $\Delta z \lesssim 0.02$, though the precise value depends on the redshift bin under consideration. We discuss possible ways to mitigate the impact of our dominant systematics in future analyses., Comment: submitted to MNRAS

We use 26 million galaxies from the Dark Energy Survey (DES) Year 1 shape catalogs over 1321 deg$^2$ of the sky to produce the most significant measurement of cosmic shear in a galaxy survey to date. We constrain cosmological parameters in both the flat $\Lambda$CDM and $w$CDM models, while also varying the neutrino mass density. These results are shown to be robust using two independent shape catalogs, two independent \photoz\ calibration methods, and two independent analysis pipelines in a blind analysis. We find a 3.5\% fractional uncertainty on $\sigma_8(\Omega_m/0.3)^{0.5} = 0.782^{+0.027}_{-0.027}$ at 68\% CL, which is a factor of 2.5 improvement over the fractional constraining power of our DES Science Verification results. In $w$CDM, we find a 4.8\% fractional uncertainty on $\sigma_8(\Omega_m/0.3)^{0.5} = 0.777^{+0.036}_{-0.038}$ and a dark energy equation-of-state $w=-0.95^{+0.33}_{-0.39}$. We find results that are consistent with previous cosmic shear constraints in $\sigma_8$ -- $\Omega_m$, and see no evidence for disagreement of our weak lensing data with data from the CMB. Finally, we find no evidence preferring a $w$CDM model allowing $w\ne -1$. We expect further significant improvements with subsequent years of DES data, which will more than triple the sky coverage of our shape catalogs and double the effective integrated exposure time per galaxy., Comment: 32 pages, 19 figures; matches PRD referee response version

We construct the largest curved-sky galaxy weak lensing mass map to date from the DES first-year (DES Y1) data. The map, about 10 times larger than previous work, is constructed over a contiguous $\approx1,500 $deg$^2$, covering a comoving volume of $\approx10 $Gpc$^3$. The effects of masking, sampling, and noise are tested using simulations. We generate weak lensing maps from two DES Y1 shear catalogs, Metacalibration and Im3shape, with sources at redshift $0.2

We describe the derivation and validation of redshift distribution estimates and their uncertainties for the galaxies used as weak lensing sources in the Dark Energy Survey (DES) Year 1 cosmological analyses. The Bayesian Photometric Redshift (BPZ) code is used to assign galaxies to four redshift bins between z=0.2 and 1.3, and to produce initial estimates of the lensing-weighted redshift distributions $n^i_{PZ}(z)$ for bin i. Accurate determination of cosmological parameters depends critically on knowledge of $n^i$ but is insensitive to bin assignments or redshift errors for individual galaxies. The cosmological analyses allow for shifts $n^i(z)=n^i_{PZ}(z-\Delta z^i)$ to correct the mean redshift of $n^i(z)$ for biases in $n^i_{\rm PZ}$. The $\Delta z^i$ are constrained by comparison of independently estimated 30-band photometric redshifts of galaxies in the COSMOS field to BPZ estimates made from the DES griz fluxes, for a sample matched in fluxes, pre-seeing size, and lensing weight to the DES weak-lensing sources. In companion papers, the $\Delta z^i$ are further constrained by the angular clustering of the source galaxies around red galaxies with secure photometric redshifts at 0.15

We present cosmological results from a combined analysis of galaxy clustering and weak gravitational lensing, using 1321 deg$^2$ of $griz$ imaging data from the first year of the Dark Energy Survey (DES Y1). We combine three two-point functions: (i) the cosmic shear correlation function of 26 million source galaxies in four redshift bins, (ii) the galaxy angular autocorrelation function of 650,000 luminous red galaxies in five redshift bins, and (iii) the galaxy-shear cross-correlation of luminous red galaxy positions and source galaxy shears. To demonstrate the robustness of these results, we use independent pairs of galaxy shape, photometric redshift estimation and validation, and likelihood analysis pipelines. To prevent confirmation bias, the bulk of the analysis was carried out while blind to the true results; we describe an extensive suite of systematics checks performed and passed during this blinded phase. The data are modeled in flat $\Lambda$CDM and $w$CDM cosmologies, marginalizing over 20 nuisance parameters, varying 6 (for $\Lambda$CDM) or 7 (for $w$CDM) cosmological parameters including the neutrino mass density and including the 457 $\times$ 457 element analytic covariance matrix. We find consistent cosmological results from these three two-point functions, and from their combination obtain $S_8 \equiv \sigma_8 (\Omega_m/0.3)^{0.5} = 0.783^{+0.021}_{-0.025}$ and $\Omega_m = 0.264^{+0.032}_{-0.019}$ for $\Lambda$CDM for $w$CDM, we find $S_8 = 0.794^{+0.029}_{-0.027}$, $\Omega_m = 0.279^{+0.043}_{-0.022}$, and $w=-0.80^{+0.20}_{-0.22}$ at 68% CL. The precision of these DES Y1 results rivals that from the Planck cosmic microwave background measurements, allowing a comparison of structure in the very early and late Universe on equal terms. Although the DES Y1 best-fit values for $S_8$ and $\Omega_m$ are lower than the central values from Planck ..., Comment: Matches published version. Results essentially unchanged, except updated covariance matrix leads to improved chi^2 (colored text removed)

We present galaxy-galaxy lensing measurements from 1321 sq. deg. of the Dark Energy Survey (DES) Year 1 (Y1) data. The lens sample consists of a selection of 660,000 red galaxies with high-precision photometric redshifts, known as redMaGiC, split into five tomographic bins in the redshift range $0.15 < z < 0.9$. We use two different source samples, obtained from the Metacalibration (26 million galaxies) and Im3shape (18 million galaxies) shear estimation codes, which are split into four photometric redshift bins in the range $0.2 < z < 1.3$. We perform extensive testing of potential systematic effects that can bias the galaxy-galaxy lensing signal, including those from shear estimation, photometric redshifts, and observational properties. Covariances are obtained from jackknife subsamples of the data and validated with a suite of log-normal simulations. We use the shear-ratio geometric test to obtain independent constraints on the mean of the source redshift distributions, providing validation of those obtained from other photo-$z$ studies with the same data. We find consistency between the galaxy bias estimates obtained from our galaxy-galaxy lensing measurements and from galaxy clustering, therefore showing the galaxy-matter cross-correlation coefficient $r$ to be consistent with one, measured over the scales used for the cosmological analysis. The results in this work present one of the three two-point correlation functions, along with galaxy clustering and cosmic shear, used in the DES cosmological analysis of Y1 data, and hence the methodology and the systematics tests presented here provide a critical input for that study as well as for future cosmological analyses in DES and other photometric galaxy surveys., Comment: 26 pages, 19 figures. Matches the version accepted by PRD

Galaxy cross-correlations with high-fidelity redshift samples hold the potential to precisely calibrate systematic photometric redshift uncertainties arising from the unavailability of complete and representative training and validation samples of galaxies. However, application of this technique in the Dark Energy Survey (DES) is hampered by the relatively low number density, small area, and modest redshift overlap between photometric and spectroscopic samples. We propose instead using photometric catalogs with reliable photometric redshifts for photo-z calibration via cross-correlations. We verify the viability of our proposal using redMaPPer clusters from the Sloan Digital Sky Survey (SDSS) to successfully recover the redshift distribution of SDSS spectroscopic galaxies. We demonstrate how to combine photo-z with cross-correlation data to calibrate photometric redshift biases while marginalizing over possible clustering bias evolution in either the calibration or unknown photometric samples. We apply our method to DES Science Verification (DES SV) data in order to constrain the photometric redshift distribution of a galaxy sample selected for weak lensing studies, constraining the mean of the tomographic redshift distributions to a statistical uncertainty of $\Delta z \sim \pm 0.01$. We forecast that our proposal can in principle control photometric redshift uncertainties in DES weak lensing experiments at a level near the intrinsic statistical noise of the experiment over the range of redshifts where redMaPPer clusters are available. Our results provide strong motivation to launch a program to fully characterize the systematic errors from bias evolution and photo-z shapes in our calibration procedure., Comment: 12 pages, 5 figures. Submitted to MNRAS. Comments welcome!

Galaxies and their dark matter halos populate a complicated filamentary network around large, nearly empty regions known as cosmic voids. Cosmic voids are usually identified in spectroscopic galaxy surveys, where 3D information about the large-scale structure of the Universe is available. Although an increasing amount of photometric data is being produced, its potential for void studies is limited since photometric redshifts induce line-of-sight position errors of $\sim50$ Mpc/$h$ or more that can render many voids undetectable. In this paper we present a new void finder designed for photometric surveys, validate it using simulations, and apply it to the high-quality photo-$z$ redMaGiC galaxy sample of the Dark Energy Survey Science Verification (DES-SV) data. The algorithm works by projecting galaxies into 2D slices and finding voids in the smoothed 2D galaxy density field of the slice. Fixing the line-of-sight size of the slices to be at least twice the photo-$z$ scatter, the number of voids found in these projected slices of simulated spectroscopic and photometric galaxy catalogs is within 20% for all transverse void sizes, and indistinguishable for the largest voids of radius $\sim 70$ Mpc/$h$ and larger. The positions, radii, and projected galaxy profiles of photometric voids also accurately match the spectroscopic void sample. Applying the algorithm to the DES-SV data in the redshift range $0.2

We study the detailed evolution of the fine-structure constant $\alpha$ in the string-inspired runaway dilaton class of models of Damour, Piazza and Veneziano. We provide constraints on this scenario using the most recent $\alpha$ measurements and discuss ways to distinguish it from alternative models for varying $\alpha$. For model parameters which saturate bounds from current observations, the redshift drift signal can differ considerably from that of the canonical $\Lambda$CDM paradigm at high redshifts. Measurements of this signal by the forthcoming European Extremely Large Telescope (E-ELT), together with more sensitive $\alpha$ measurements, will thus dramatically constrain these scenarios., Comment: 11 pages, 4 figures

Very long baseline interferometry at millimetre/submillimetre wavelengths (mmVLBI) offers the highest achievable spatial resolution at any wavelength in astronomy. The anticipated inclusion of ALMA as a phased array into a global VLBI network will bring unprecedented sensitivity and a transformational leap in capabilities for mmVLBI. Building on years of pioneering efforts in the US and Europe the ongoing ALMA Phasing Project (APP), a US-led international collaboration with MPIfR-led European contributions, is expected to deliver a beamformer and VLBI capability to ALMA by the end of 2014 (APP: Fish et al. 2013, arXiv:1309.3519). This report focuses on the future use of mmVLBI by the international users community from a European viewpoint. Firstly, it highlights the intense science interest in Europe in future mmVLBI observations as compiled from the responses to a general call to the European community for future research projects. A wide range of research is presented that includes, amongst others: - Imaging the event horizon of the black hole at the centre of the Galaxy - Testing the theory of General Relativity an/or searching for alternative theories - Studying the origin of AGN jets and jet formation - Cosmological evolution of galaxies and BHs, AGN feedback - Masers in the Milky Way (in stars and star-forming regions) - Extragalactic emission lines and astro-chemistry - Redshifted absorption lines in distant galaxies and study of the ISM and circumnuclear gas - Pulsars, neutron stars, X-ray binaries - Testing cosmology - Testing fundamental physical constants, Comment: Replaced figures 2 and 3: corrected position SRT. Corrected minor typo in 5.1

Various classes of exotic singularity models have been studied as possible mimic models for the observed recent acceleration of the universe. Here we further study one of these classes and, under the assumption that they are phenomenological toy models for the behavior of an underlying scalar field which also couples to the electromagnetic sector of the theory, obtain the corresponding behavior of the fine-structure constant $\alpha$ for particular choices of model parameters that have been previously shown to be in reasonable agreement with cosmological observations. We then compare this predicted behavior with available measurements of $\alpha$, thus constraining this putative coupling to electromagnetism. We find that values of the coupling which would provide a good fit to spectroscopic measurements of $\alpha$ are in more than three-sigma tension with local atomic clock bounds. Future measurements by ESPRESSO and ELT-HIRES will provide a definitive test of these models., Comment: 8 pages, 12 figures, 3 tables, REVTEX 4-1

We discuss how dynamical dark energy universes with exotic singularities may be distinguished from the standard $\Lambda$CDM model on the basis of their redshift drift signal, for which measurements both in the acceleration phase and in the deep matter era will be provided by forthcoming astrophysical facilities. Two specific classes of exotic singularity models are studied: sudden future singularity models and finite scale factor singularity models. In each class we identify the models which can mimic $\Lambda$CDM and play the role of dark energy as well as models for which redshift drift signals are significantly different from $\Lambda$CDM and the test can differentiate between them., Comment: 6 pages, 8 figures, Revtex 4-1, references added

Beyond the standard cosmological model the late-time accelerated expansion of the universe can be reproduced by the introduction of an additional dynamical scalar field. In this case, the field is expected to be naturally coupled to the rest of the theory's fields, unless a (still unknown) symmetry suppresses this coupling. Therefore, this would possibly lead to some observational consequences, such as space-time variations of nature's fundamental constants. In this paper we investigate the coupling between a dynamical Dark Energy model and the electromagnetic field, and the corresponding evolution of the fine structure constant ($\alpha$) with respect to the standard local value $\alpha_0$. In particular, we derive joint constraints on two dynamical Dark Energy model parametrizations (the Chevallier-Polarski-Linder and Early Dark Energy model) and on the coupling with electromagnetism $\zeta$, forecasting future low-medium redshift observations. We combine supernovae and weak lensing measurements from the Euclid experiment with high-resolution spectroscopy measurements of fundamental couplings and the redshift drift from the European Extremely Large Telescope, highlighting the contribution of each probe. Moreover, we also consider the case where the field driving the $\alpha$ evolution is not the one responsible for cosmic acceleration and investigate how future observations can constrain this scenario., Comment: 15 pages, 12 figures

The observational possibilities enabled by an ultra-stable CODEX-like spectrograph at the E-ELT will open new doors in the characterisation of the nature of Dark Energy. Indeed, it will provide measurements of a so far unexplored redshift-range ($2

The recent observational evidence for the acceleration of the universe demonstrates that canonical theories of cosmology and particle physics are incomplete - if not incorrect - and that new physics is out there, waiting to be discovered. The most exciting task for the next generation of astrophysical facilities is therefore to search for, identify and ultimately characterise this new physics. Here we summarise ongoing work by CAUP's Dark Side Team aiming to identify optimal observational strategies for forthcoming facilities. The discussion is centred around the E-ELT (which will play a key role in this endeavour), but synergies with other ground and space-based facilities are also briefly considered. With the expected major improvements in the sensitivity of standard probes and entirely new ones such as the redshift drift (which the E-ELT, the SKA and possibly other facilities will measure) a new generation of precision consistency tests of the standard paradigm will become possible., Comment: To appear in the Proceedings of the Sesto Conference on Varying Fundamental Constants and Dynamical Dark Energy (Sesto, Italy, July 8-13, 2013)

We study cosmological models involving scalar fields coupled to radiation and discuss their effect on the redshift evolution of the cosmic microwave background temperature, focusing on links with varying fundamental constants and dynamical dark energy. We quantify how allowing for the coupling of scalar fields to photons, and its important effect on luminosity distances, weakens current and future constraints on cosmological parameters. In particular, for evolving dark energy models, joint constraints on the dark energy equation of state combining BAO radial distance and SN luminosity distance determinations, will be strongly dominated by BAO. Thus, to fully exploit future SN data one must also independently constrain photon number non-conservation arising from the possible coupling of SN photons to the dark energy scalar field. We discuss how observational determinations of the background temperature at different redshifts can, in combination with distance measures data, set tight constraints on interactions between scalar fields and photons, thus breaking this degeneracy. We also discuss prospects for future improvements, particularly in the context of Euclid and the E-ELT and show that Euclid can, even on its own, provide useful dark energy constraints while allowing for photon number non-conservation., Comment: 29 pages, 11 figures; v2: Matches published version (JCAP)

Future redshift-drift measurements (also known as Sandage-Loeb signal) will be crucial to probe the so called "redshift desert", thus providing a new tool for cosmological studies. In this paper we quantify the ability of a future measurement of the Sandage-Loeb signal by a CODEX-like spectrograph to constrain a phenomenological parametrization of dynamical dark energy, specifically by obtaining constraints on $w_0$ and $w_a$. We also demonstrate that if used alongside CMB data, the Sandage-Loeb measurements will be able to break degeneracies between expansion parameters, thus greatly improving cosmological constraints., Comment: 6 pages, 7 figures. New version matching the one accepted by PRD

Many cosmological models invoke rolling scalar fields to account for the observed acceleration of the expansion of the universe. These theories generally include a potential V(phi) which is a function of the scalar field phi. Although V(phi) can be represented by a very diverse set of functions, recent work has shown the under some conditions, such as the slow roll conditions, the equation of state parameter w is either independent of the form of V(phi) or is part of family of solutions with only a few parameters. In realistic models of this type the scalar field couples to other sectors of the model leading to possibly observable changes in the fundamental constants such as the fine structure constant alpha and the proton to electron mass ratio mu. This paper explores the limits this puts on the validity of various cosmologies that invoke rolling scalar fields. We find that the limit on the variation of mu puts significant constraints on the product of a cosmological parameter w+1 times a new physics parameter zeta_mu^2, the coupling constant between mu and the rolling scalar field. Even when the cosmologies are restricted to very slow roll conditions either the value of zeta_mu must be at the lower end of or less than its expected values or the value of w+1 must be restricted to values vanishingly close to 0. This implies that either the rolling scalar field is very weakly coupled with the electromagnetic field, small zeta_mu, very weakly coupled with gravity, w+1 ~ 0 or both. These results stress that adherence to the measured invariance in mu is a very significant test of the validity of any proposed cosmology and any new physics it requires. The limits on the variation of mu also produces a significant tension with the reported changes in the value of alpha., Comment: Accepted for publication in MNRAS 10 pages, 6 figures

Precision measurements of nature's fundamental couplings and a first measurement of the cosmological redshift drift are two of the key targets for future high-resolution ultra-stable spectrographs such as CODEX. Being able to do both gives CODEX a unique advantage, allowing it to probe dynamical dark energy models (by measuring the behavior of their equation of state) deep in the matter era and thereby testing classes of models that would otherwise be difficult to distinguish from the standard $\Lambda$CDM paradigm. We illustrate this point with two simple case studies., Comment: 4 pages, 4 figures; submitted to Phys. Rev. D