arXiv subjects

François Bouchet

Publications and source records attributed to François Bouchet.

12 recordsLinked to original sources

arXiv2024

A Fair Post-Processing Method based on the MADD Metric for Predictive Student Models

Predictive student models are increasingly used in learning environments. However, due to the rising social impact of their usage, it is now all the more important for these models to be both sufficiently accurate and fair in their predictions. To evaluate algorithmic fairness, a new metric has been developed in education, namely the Model Absolute Density Distance (MADD). This metric enables us to measure how different a predictive model behaves regarding two groups of students, in order to quantify its algorithmic unfairness. In this paper, we thus develop a post-processing method based on this metric, that aims at improving the fairness while preserving the accuracy of relevant predictive models' results. We experiment with our approach on the task of predicting student success in an online course, using both simulated and real-world educational data, and obtain successful results. Our source code and data are in open access at https://github.com/melinaverger/MADD .

cs.CY↗

arXiv2023

Is Your Model "MADD"? A Novel Metric to Evaluate Algorithmic Fairness for Predictive Student Models

Predictive student models are increasingly used in learning environments due to their ability to enhance educational outcomes and support stakeholders in making informed decisions. However, predictive models can be biased and produce unfair outcomes, leading to potential discrimination against some students and possible harmful long-term implications. This has prompted research on fairness metrics meant to capture and quantify such biases. Nonetheless, so far, existing fairness metrics used in education are predictive performance-oriented, focusing on assessing biased outcomes across groups of students, without considering the behaviors of the models nor the severity of the biases in the outcomes. Therefore, we propose a novel metric, the Model Absolute Density Distance (MADD), to analyze models' discriminatory behaviors independently from their predictive performance. We also provide a complementary visualization-based analysis to enable fine-grained human assessment of how the models discriminate between groups of students. We evaluate our approach on the common task of predicting student success in online courses, using several common predictive classification models on an open educational dataset. We also compare our metric to the only predictive performance-oriented fairness metric developed in education, ABROCA. Results on this dataset show that: (1) fair predictive performance does not guarantee fair models' behaviors and thus fair outcomes, (2) there is no direct relationship between data bias and predictive performance bias nor discriminatory behaviors bias, and (3) trained on the same data, models exhibit different discriminatory behaviors, according to different sensitive features too. We thus recommend using the MADD on models that show satisfying predictive performance, to gain a finer-grained understanding on how they behave and to refine models selection and their usage.

cs.LG↗

arXiv2021

Breaking the degeneracy between polarization efficiency and cosmological parameters in CMB experiments

Accurate cosmological parameter estimates using polarization data of the cosmic microwave background (CMB) put stringent requirements on map calibration, as highlighted in the recent results from the Planck satellite. In this paper, we point out that a model-dependent determination of polarization calibration can be achieved by the joint fit of the TE and EE CMB power spectra. This provides a valuable cross-check to band-averaged polarization efficiency measurements determined using other approaches. We demonstrate that, in $Λ$CDM, the combination of the TE and EE constrain polarization calibration with sub-percent uncertainty with Planck data and 2% uncertainty with SPTpol data. We arrive at similar conclusions when extending $Λ$CDM to include the amplitude of lensing $A_{\rm L}$, the number of relativistic species $N_{\rm eff}$, or the sum of the neutrino masses $\sum m_ν$. The uncertainties on cosmological parameters are minimally impacted when marginalizing over polarization calibration, except, as can be expected, for the uncertainty on the amplitude of the primordial scalar power spectrum $\ln(10^{10} A_{\rm s})$, which increases by $20-50$%. However, this information can be fully recovered by adding TT data. For current and future ground-based experiments, SPT-3G and CMB-S4, we forecast the cosmological parameter uncertainties to be minimally degraded when marginalizing over polarization calibration parameters. In addition, CMB-S4 could constrain its polarization calibration at the level of $\sim$0.2% by combining TE and EE, and reach $\sim$0.06% by also including TT. We therefore conclude that relying on calibrating against Planck polarization maps, whose statistical uncertainty is limited to $\sim$0.5%, would be insufficient for upcoming experiments.

astro-ph.CO↗

arXiv2019

PICO: Probe of Inflation and Cosmic Origins

The Probe of Inflation and Cosmic Origins (PICO) is an imaging polarimeter that will scan the sky for 5 years in 21 frequency bands spread between 21 and 799 GHz. It will produce full-sky surveys of intensity and polarization with a final combined-map noise level of 0.87 $μ$K arcmin for the required specifications, equivalent to 3300 Planck missions, and with our current best-estimate would have a noise level of 0.61 $μ$K arcmin (6400 Planck missions). PICO will either determine the energy scale of inflation by detecting the tensor to scalar ratio at a level $r=5\times 10^{-4}~(5σ)$, or will rule out with more than $5σ$ all inflation models for which the characteristic scale in the potential is the Planck scale. With LSST's data it could rule out all models of slow-roll inflation. PICO will detect the sum of neutrino masses at $>4σ$, constrain the effective number of light particle species with $ΔN_{\rm eff}<0.06~(2σ)$, and elucidate processes affecting the evolution of cosmic structures by measuring the optical depth to reionization with errors limited by cosmic variance and by constraining the evolution of the amplitude of linear fluctuations $σ_{8}(z)$ with sub-percent accuracy. Cross-correlating PICO's map of the thermal Sunyaev-Zeldovich effect with LSST's gold sample of galaxies will precisely trace the evolution of thermal pressure with $z$. PICO's maps of the Milky Way will be used to determine the make up of galactic dust and the role of magnetic fields in star formation efficiency. With 21 full sky legacy maps in intensity and polarization, which cannot be obtained in any other way, the mission will enrich many areas of astrophysics. PICO is the only single-platform instrument with the combination of sensitivity, angular resolution, frequency bands, and control of systematic effects that can deliver this compelling, timely, and broad science.

astro-ph.IM↗

arXiv2018

Cosmic Microwave Background Constraints in Light of Priors Over Reionization Histories

Non-parametric reconstruction or marginalization over the history of reionization using cosmic microwave background data necessarily assumes a prior over possible histories. We show that different but reasonable choices of priors can shift current and future constraints on the reionization optical depth, $τ$, or correlated parameters such as the neutrino mass sum, $Σm_ν$, at the level of 0.3-0.4$σ$, i.e., that this analysis is somewhat prior dependent. We point out some prior-related problems with the commonly used principal component reconstruction, concluding that the significance of some recent hints of early reionization in Planck 2015 data has been overestimated. We also present the first non-parametric reconstruction applied to newer Planck intermediate (2016) data and find that the hints of early reionization disappear entirely in this more precise dataset. These results limit possible explanations of the EDGES 21cm signal which would have also significantly reionized the universe at $z\,{>}\,15$. Our findings about the dependence on priors motivate the pursuit of improved data or searches for physical reionization models which can reduce the prior volume. The discussion here of priors is of general applicability to other non-parametric reconstructions, for example of the primordial power spectrum, of the recombination history, or of the expansion rate.

astro-ph.CO↗

arXiv2017

Exploring cosmic origins with CORE: gravitational lensing of the CMB

Lensing of the CMB is now a well-developed probe of large-scale clustering over a broad range of redshifts. By exploiting the non-Gaussian imprints of lensing in the polarization of the CMB, the CORE mission can produce a clean map of the lensing deflections over nearly the full-sky. The number of high-S/N modes in this map will exceed current CMB lensing maps by a factor of 40, and the measurement will be sample-variance limited on all scales where linear theory is valid. Here, we summarise this mission product and discuss the science that it will enable. For example, the summed mass of neutrinos will be determined to an accuracy of 17 meV combining CORE lensing and CMB two-point information with contemporaneous BAO measurements, three times smaller than the minimum total mass allowed by neutrino oscillations. In the search for B-mode polarization from primordial gravitational waves with CORE, lens-induced B-modes will dominate over instrument noise, limiting constraints on the gravitational wave power spectrum amplitude. With lensing reconstructed by CORE, one can "delens" the observed polarization internally, reducing the lensing B-mode power by 60%. This improves to 70% by combining lensing and CIB measurements from CORE, reducing the error on the gravitational wave amplitude by 2.5 compared to no delensing (in the null hypothesis). Lensing measurements from CORE will allow calibration of the halo masses of the 40000 galaxy clusters that it will find, with constraints dominated by the clean polarization-based estimators. CORE can accurately remove Galactic emission from CMB maps with its 19 frequency channels. We present initial findings that show that residual Galactic foreground contamination will not be a significant source of bias for lensing power spectrum measurements with CORE. [abridged]

astro-ph.CO↗

arXiv2014

Dancing in the dark: galactic properties trace spin swings along the cosmic web

A large-scale hydrodynamical cosmological simulation, Horizon-AGN, is used to investigate the alignment between the spin of galaxies and the cosmic filaments above redshift 1.2. The analysis of more than 150 000 galaxies per time step in the redshift range 1.2<z<1.8 with morphological diversity shows that the spin of low-mass blue galaxies is preferentially aligned with their neighbouring filaments, while high-mass red galaxies tend to have a perpendicular spin. The reorientation of the spin of massive galaxies is provided by galaxy mergers, which are significant in their mass build-up. We find that the stellar mass transition from alignment to misalignment happens around 3.10^10 M_sun. Galaxies form in the vorticity-rich neighbourhood of filaments, and migrate towards the nodes of the cosmic web as they convert their orbital angular momentum into spin. The signature of this process can be traced to the properties of galaxies, as measured relative to the cosmic web. We argue that a strong source of feedback such as active galactic nuclei is mandatory to quench in situ star formation in massive galaxies and promote various morphologies. It allows mergers to play their key role by reducing post-merger gas inflows and, therefore, keeping spins misaligned with cosmic filaments.

astro-ph.CO↗

arXiv2014

PRISM (Polarized Radiation Imaging and Spectroscopy Mission): An Extended White Paper

PRISM (Polarized Radiation Imaging and Spectroscopy Mission) was proposed to ESA in May 2013 as a large-class mission for investigating within the framework of the ESA Cosmic Vision program a set of important scientific questions that require high resolution, high sensitivity, full-sky observations of the sky emission at wavelengths ranging from millimeter-wave to the far-infrared. PRISM's main objective is to explore the distant universe, probing cosmic history from very early times until now as well as the structures, distribution of matter, and velocity flows throughout our Hubble volume. PRISM will survey the full sky in a large number of frequency bands in both intensity and polarization and will measure the absolute spectrum of sky emission more than three orders of magnitude better than COBE FIRAS. The aim of this Extended White Paper is to provide a more detailed overview of the highlights of the new science that will be made possible by PRISM

astro-ph.CO↗

arXiv2014

CMB Polarization can constrain cosmology better than CMB temperature

We demonstrate that for a cosmic variance limited experiment, CMB E polarization alone places stronger constraints on cosmological parameters than CMB temperature. For example, we show that EE can constrain parameters better than TT by up to a factor 2.8 when a multipole range of l=30-2500 is considered. We expose the physical effects at play behind this remarkable result and study how it depends on the multipole range included in the analysis. In most relevant cases, TE or EE surpass the TT based cosmological constraints. This result is important as the small scale astrophysical foregrounds are expected to have a much reduced impact on polarization, thus opening the possibility of building cleaner and more stringent constraints of the LCDM model. This is relevant specially for proposed future CMB satellite missions, such as CORE or PRISM, that are designed to be cosmic variance limited in polarization till very large multipoles. We perform the same analysis for a Planck-like experiment, and conclude that even in this case TE alone should determine the constraint on $Ω_ch^2$ better than TT by 15%, while determining $Ω_bh^2$, $n_s$ and $θ$ with comparable accuracy. Finally, we explore a few classical extensions of the LCDM model and show again that CMB polarization alone provides more stringent constraints than CMB temperature in case of a cosmic variance limited experiment.

astro-ph.CO↗

arXiv2013

Angular Correlation Functions for Models with Logarithmic Oscillations

There exist several theoretical motivations for primordial correlation functions (such as the power spectrum) to contain oscillations as a logarithmic function of comoving momentum k. While these features are commonly searched for in k-space, an alternative is to use angular space; that is, search for correlations between the directional vectors of observation. We develop tools to efficiently compute the angular correlations based on a stationary phase approximation and examine several example oscillations in the primordial power spectrum, bispectrum, and trispectrum. We find that logarithmically-periodic oscillations are essentially featureless and therefore difficult to detect using the standard correlator, though others might be feasible.

hep-th↗

arXiv2012

Non Gaussianity and Minkowski Functionals: forecasts for Planck

We study Minkowski Functionals as probes of primordial non-Gaussianity in the Cosmic Microwave Background, specifically for the estimate of the primordial `local' bi-spectrum parameter f_NL, with instrumental parameters which should be appropriate for the Planck experiment. We use a maximum likelihood approach, which we couple with various filtering methods and test thoroughly for convergence. We included the effect of inhomogeneous noise as well as astrophysical biases induced by point sources and by the contamination from the Galaxy. We find that, when Wiener filtered maps are used (rather than simply smoothed with Gaussian), the expected error on the measurement of f_NL should be as small as Δf_NL \simeq 10 when combining the 3 channels at 100, 143 and 217 GHz in the Planck extended mission setup. This result is fairly insensitive to the non homogeneous nature of the noise, at least for realistic hit-maps expected from Planck. We then estimate the bias induced on the measurement of f_NL by point sources in those 3 channels. With the appropriate masking of the bright sources, this bias can be reduced to a negligible level in the 100 and 143 GHz channels. It remains significant in the 217 GHz channel, but can be corrected for. The galactic foreground biases are quite important and present a complex dependence on sky coverage: making them negligible will depend strongly on the quality of the component separation methods.

astro-ph.CO↗

arXiv2003

Gaussianity of Cosmic Velocity Fields and Linearity of the Velocity-Gravity Relation

We present a numerical study of the relation between the cosmic peculiar velocity field and the gravitational acceleration field. We show that on mildly non-linear scales (4-10 Mpc Gaussian smoothing), the distribution of the Cartesian coordinates of each of these fields is well approximated by a Gaussian. In particular, their kurtoses and negentropies are small compared to those of the velocity divergence and density fields. We find that at these scales the relation between the velocity and gravity field follows linear theory to good accuracy. Specifically, the systematic errors in velocity-velocity comparisons due to assuming the linear model do not exceed 6% in beta. To correct for them, we test various nonlinear estimators of velocity from density. We show that a slight modification of the alpha-formula proposed by Kudlicki et al. yields an estimator which is essentially unbiased and has a small variance.

astro-ph↗