DCDavid Cornu

David Cornu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


David Cornu has a total of 165 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Modeling the 3D Milky Way using Machine Learning with Gaia and infrared surveys

    by David Cornu

    Modélisation de la voie lactée en 3D par machine learning avec les données infrarouges et Gaia: La structure à grande échelle de la Voie-Lactée (VL) n'est actuellement toujours pas parfaitement contrainte. Contrairement aux autres galaxies, il est difficile d'observer directement sa structure du fait de notre appartenance à celle-ci. La confusion entre les étoiles et l'occultation de la lumière par le milieu interstellaire (MIS) sont les principales sources de difficulté qui empêchent la reconstruction de la structure sous-jacente de la VL. Par ailleurs, de plus en plus de relevés astronomiques de grande ampleur sont disponibles et permettent de surmonter ces difficultés. Le relevé Gaia et ses 1.6 milliards mesures de distances aux étoiles est le nouvel outil de prédilection pour l’étude de la structure de la VL et l’analyse des populations stellaires. Ces nouvelles données peuvent être combinées avec d’autres grands relevés infrarouges (IR) afin d’effectuer des mesures à des distances jusque-là inégalées. Par ailleurs, le nombre d’applications reposant sur des méthodes d’apprentissage machine (AM) s’est envolé ces vingt dernières années et celles-ci sont de plus en plus employées en astronomie. Ces méthodes sont capables d’automatiser la résolution de problèmes complexes ou encore d’extraire efficacement des statistiques sur de grands jeux de données.Dans cette étude, nous commençons par décrire la construction d’un outil de classification par AM utilisé pour améliorer les méthodes classiques de classification des Jeunes Objets Stellaires (JOS). Comme les étoiles naissent dans un environnement interstellaire dense, il est possible d’utiliser les plus jeunes d’entre elles, qui n’ont pas encore eu le temps de s’éloigner de leur lieux de formation, afin d’identifier les structures denses du MIS. La combinaison des JOS et des distances mesurées par Gaia permet alors de reconstruire la structure 3D des nuages denses. Notre méthode de classification par AM est basée sur les réseaux de neurones artificiels et se sert des données du télescope spatial Spitzer pour reconstruire automatiquement la classification des JOS sur la base d’une liste d’exemples. Nous détaillons la construction des jeux de données associés ainsi que l’effet du déséquilibre entre les classes, ce qui permet d’optimiser les prédictions du réseau et d’estimer la précision associée. Cette méthode est capable d’identifier des JOS dans de très grands relevés tout en fournissant une probabilité d’appartenance pour chacun des objets testés. Celle-ci peut alors être utilisée pour retenir les objets les plus fiables afin de reconstruire la structure des nuages.Dans une seconde partie, nous présentons une méthode permettant de reconstruire la distribution 3D de l’extinction dans la VL et reposant sur des réseaux de neurones convolutifs. Cette approche permet de prédire des profils d’extinction sur la base de données IR provenant du relevé 2MASS. Ce réseau est entraîné à l’aide du modèle de la Galaxie de Besançon afin de reproduire la distribution en distance de l’extinction à grande échelle en s’appuyant sur la comparaison entre le modèle et les données observées. Nous avons ainsi reconstruit une grande portion du plan Galactique dans la région du bras de la Carène, et avons montré que notre prédiction est compétitive avec d’autres cartes d’extinction 3D qui font référence. Nos résultats sont notamment capables de prédire des structures spatialement cohérentes, et parviennent à réduire les artefacts fréquents dits ``doigts de Dieu''. Cette méthode est parvenue à résoudre des structures distantes jusqu’à 10 kpc avec une résolution formelle de 100 pc. Notre réseau est également capable de combiner les données 2MASS et Gaia sans avoir recours à une identification croisée. Cela permet d’utiliser automatiquement le jeu de données le plus pertinent en fonction de la distance. Les résultats de cette prédiction combinée sont encourageants et ouvrent la voie à de nouvelles reconstructions du plan Galactique en combinant davantage de jeux de données.

  • Open Access

    YOLO-CIANNA: Galaxy detection with deep learning in radio data

    by D. Cornu, B. Semelin, P. Salomé, X. Lu, S. Aicardi, J. Freundlich, F. Mertens, A. Marchal, G. Sainton, F. Combes, et al.

    Context . As the scientific exploitation of the Square Kilometre Array (SKA) approaches, there is a need for new advanced data analysis and visualization tools capable of processing large high-dimensional datasets. Aims . In this study, we aim to generalize the YOLO-CIANNA deep learning source detection and characterization method for 3D hyperspectral HI emission cubes. Methods . We present the adaptations we made to the regression-based detection formalism and the construction of an end-to-end 3D convolutional neural network (CNN) backbone. We then describe a processing pipeline for applying the method to simulated 3D HI cubes from the SKA Observatory Science Data Challenge 2 (SDC2) dataset. Results . The YOLO-CIANNA method was originally developed and used by the MINERVA team that won the official SDC2 competition. Despite the public release of the full SDC2 dataset, no published result has yet surpassed MINERVA’s top score. In this paper, we present an updated version of our method that improves our challenge score by 9.5%. The resulting catalog exhibits a high detection purity of 92.3%, best-in-class characterization accuracy, and contains 45% more confirmed sources than concurrent classical detection tools. The method is also computationally efficient, processing the full ~1TB SDC2 data cube in 30 min on a single GPU. Conclusions . These state-of-the-art results highlight the effectiveness of 3D CNN-based detectors for processing large hyperspectral data cubes and represent a promising step toward applying YOLO-CIANNA to observational data from SKA and its precursors.


  • Open Access

    A new dust map of the Milky Way

    by D. J. Marshall, J. Montillaud, L. Cambrésy, D. Cornu

    Aims . We present REDLINE, a new technique for mapping three-dimensional interstellar extinction across the Galactic plane, with the aim of revealing the distribution of interstellar dust at distances where traditional mapping approaches face limitations, particularly beyond 4 kpc from the Sun. Methods . Our approach relies solely on near-infrared stellar observations to achieve a higher sensitivity through high-extinction regions. Building on our previous mapping techniques, REDLINE employs Markov chain Monte Carlo (MCMC) methods to constrain the line-of-sight dust distribution. A further methodological improvement is the incorporation of an enhanced Besançon Galaxy model (BGM), which provides more accurate stellar population predictions. We applied this refined framework to survey data covering the entire Galactic plane at | b | ≤ 1 ° , resulting in a map with 10 ′ angular resolution but on a 5 ′ grid and with a 100 pc resolution in distance. Results . REDLINE successfully maps the structure of the Galactic disc, particularly in the inner Galaxy. While showing general agreement with existing maps of the large-scale dust distribution, our method provides superior sensitivity at greater distances compared to star-by-star analyses that excel at closer ranges. We identified several key spiral arm features and provide new constraints on dust distribution in previously poorly mapped regions. A significant finding from the REDLINE map indicates that dust in the Galactic disc is distributed in a highly non-uniform manner, revealing a fragmented structure with clumps of matter aligning along expected spiral patterns, rather than continuous spiral arms, and highlighting regions of reduced density that deviate from idealised logarithmic models. The seamless and extended view provided by REDLINE demonstrates its value for advancing understanding of the Milky Way’s interstellar medium and spiral structure.


  • Open Access

    Identification of molecular line emission using convolutional neural networks

    by N. Kessler, T. Csengeri, D. Cornu, S. Bontemps, L. Bouscasse

    Context. Complex organic molecules (COMs) are found to be abundant in various astrophysical environments, particularly toward star-forming regions, where they are observed both toward protostellar envelopes as well as shocked regions. The emission spectrum, especially that of heavier COMs, might consist of up to hundreds of lines, where line blending hinders the analysis. However, identifying the molecular composition of the gas that leads to the observed millimeter spectra is the first step toward a quantitative analysis. Aims. We have developed a new method based on supervised machine learning to recognize spectroscopic features of the rotational spectrum of molecules in the 3 mm atmospheric transmission band for a list of species including COMs, with the aim of obtaining a detection probability. Methods. We used local thermodynamic equilibrium (LTE) modeling to build a large set of synthetic spectra of 20 molecular species, including COMs with a range of physical conditions typical for star-forming regions. We successfully designed and trained a convolutional neural network (CNN) that provides detection probabilities of individual species in the spectra. Results. We demonstrate that the CNN model we developed has a robust performance to detect spectroscopic signatures from these species in synthetic spectra. We evaluated its ability to detect molecules according to the noise level, frequency coverage, and line-richness, as well as to test its performance for an incomplete frequency coverage with high detection probabilities for the tested parameter space, with no false predictions. Finally, we applied the CNN model to obtain predictions on observational data from the literature toward line-rich hot core-like sources, where the detection probabilities remain reasonable, with no false detections. Conclusions. We demonstrate the use of CNNs in facilitating the analysis of complex millimeter spectra both on synthetic spectra, along with the first tests performed on observational data. Further analyses on its explainability, as well as calibration using a larger observational dataset, will help improve the performance of our method for future applications.


  • Open Access

    Comparison of Bayesian inference methods using the LORELI II database of hydro-radiative simulations of the 21-cm signal

    by R. Meriot, B. Semelin, D. Cornu

    While the observation of the 21-cm signal from the cosmic dawn and epoch of reionisation is an instrumental challenge, the interpretation of a prospective detection is still open to questions regarding the modelling of the signal and the Bayesian inference techniques that bridge the gap between theory and observations. To address some of these questions, we present Loreli II, a database of nearly 10 000 simulations of the 21-cm signal run with the Licorice 3D radiative transfer code. With Loreli II, we explored a five-dimensional astrophysical parameter space where star formation, X-ray emissions, and UV emissions are varied. We then used this database to train neural networks and perform Bayesian inference on 21-cm power spectra affected by thermal noise at the level of 100 hours of observation with the Square Kilometre Array. We studied and compared three inference techniques: an emulator of the power spectrum, a neural density estimator that fits the implicit likelihood of the model, and a Bayesian neural network that directly fits the posterior distribution. We measured the performance of each method by comparing them on a statistically representative set of inferences, notably using the principles of simulation-based calibration. We report errors on the 1D marginalised posteriors (biases and over or under confidence) below $15 %$ of the standard deviation for the emulator and below $25 %$ for the other methods. We conclude that at our noise level and our sampling density of the parameter space, an explicit Gaussian likelihood is sufficient. This may not be the case at a lower noise level or if a denser sampling is used to reach higher accuracy. We then applied the emulator method to recent HERA upper limits and report weak constraints on the X-ray emissivity parameter of our model.


  • Open Access

    Combining summary statistics with simulation-based inference for the 21 cm signal from the Epoch of Reionisation

    by B. Semelin, R. Mériot, A. Mishra, D. Cornu

    The 21 cm signal from the Epoch of Reionisation will be observed with the upcoming Square Kilometer Array (SKA). We expect it to yield a full tomography of the signal, which opens up the possibility to explore its non-Gaussian properties. This raises the question of how can we extract the maximum information from tomography and derive the tightest constraint on the signal. In this work, instead of looking for the most informative summary statistics, we investigate how to combine the information from two sets of summary statistics using simulation-based inference. To this end, we trained neural density estimators (NDE) to fit the implicit likelihood of our model, the LICORICE code, using the Loreli II database. We trained three different NDEs: one to perform Bayesian inference on the power spectrum, one to perform it on the linear moments of the pixel distribution function (PDF), and one to work with the combination of the two. We performed ∼ 900 inferences at different points in our parameter space and used them to assess both the validity of our posteriors with a simulation-based calibration (SBC) and the typical gain obtained by combining summary statistics. We find that our posteriors are biased by no more than ∼ 20 % of their standard deviation and under-confident by no more than ∼ 15 %. Then, we established that combining summary statistics produces a contraction of the 4D volume of the posterior (derived from the generalised variance) in $91.5 %$ of our cases, and in $70$ to $80 %$ of the cases for the marginalised 1D posteriors. The median volume variation is a contraction of a factor of a few for the 4D posteriors and a contraction of 20 to 30 $%$ in the case of the marginalised 1D posteriors. This shows that our approach is a possible alternative to looking for so-called sufficient statistics in the theoretical sense.


  • Open Access

    The Galaxy Activity, Torus, and Outflow Survey (GATOS)

    by R. Poitevineau, F. Combes, S. Garcia-Burillo, D. Cornu, A. Alonso Herrero, C. Ramos Almeida, A. Audibert, E. Bellocchi, P. G. Boorman, A. J. Bunker, et al.

    The detailed feeding and feedback mechanisms of active galactic nuclei (AGNs) are not yet well known. For low-luminosity AGNs, obscured AGNs, and late-type galaxies, the masses of their central black holes (BH) are difficult to determine precisely. Our goal with the GATOS sample is to study the circum-nuclear regions and, in the present work, to better determine their BH mass, with more precise and accurate estimations than those obtained from scaling relations. We used the high spatial resolution of ALMA to resolve the CO(3–2) emission within ∼100 pc around the supermassive black hole (SMBH) of seven GATOS galaxies and try to estimate their BH mass when enough gas is present in the nuclear regions. We studied the seven bright (LAGN(14 − 150 keV)≥1042 erg/s) and nearby (< 28 Mpc) galaxies from the GATOS core sample. For the sake of comparison, we first searched the literature for previous BH mass estimations. We also made additional estimations using the MBH–σ relation and the fundamental plane of BH activity. We developed a new method using supervised machine learning to estimate the BH mass either from position-velocity diagrams or from first-moment maps computed from ALMA CO(3–2) observations. We used numerical simulations with a large range of parameters to create the training, validation, and test sets. Seven galaxies had sufficient gas detected, thus, we were able to make a BH estimation from the ALMA data: NGC 4388, NGC 5506, NGC 5643, NGC 6300, NGC 7314, NGC 7465, and NGC 7582. Our BH masses range from 6.39 to 7.18 log(MBH/M⊙) and are consistent with the previous estimations. In addition, our machine learning method has the advantage of providing a robust estimation of errors with confidence intervals. The method has also more growth potential than scaling relations. This work represents the first step toward an automatized method for estimating MBH using machine learning.


  • Open Access

    YOLO-CIANNA: Galaxy detection with deep learning in radio data

    by D. Cornu, P. Salomé, B. Semelin, A. Marchal, J. Freundlich, S. Aicardi, X. Lu, G. Sainton, F. Mertens, F. Combes, et al.

    Context. The upcoming Square Kilometer Array (SKA) will set a new standard regarding data volume generated by an astronomical instrument, which is likely to challenge widely adopted data-analysis tools that scale inadequately with the data size.Aims. The aim of this study is to develop a new source detection and characterization method for massive radio astronomical datasets based on modern deep-learning object detection techniques. For this, we seek to identify the specific strengths and weaknesses of this type of approach when applied to astronomical data.Methods. We introduce YOLO-CIANNA, a highly customized deep-learning object detector designed specifically for astronomical datasets. In this paper, we present the method and describe all the elements introduced to address the specific challenges of radio astronomical images. We then demonstrate the capabilities of this method by applying it to simulated 2D continuum images from the SKA observatory Science Data Challenge 1 (SDC1) dataset.Results. Using the SDC1 metric, we improve the challenge-winning score by +139% and the score of the only other post-challenge participation by +61%. Our catalog has a detection purity of 94% while detecting 40–60% more sources than previous top-score results, and exhibits strong characterization accuracy. The trained model can also be forced to reach 99% purity in post-process and still detect 10–30% more sources than the other top-score methods. It is also computationally efficient, with a peak prediction speed of 500 images of 512×512 pixels per second on a single GPU.Conclusions. YOLO-CIANNA achieves state-of-the-art detection and characterization results on the simulated SDC1 dataset and is expected to transfer well to observational data from SKA precursors.


  • SKA Science Data Challenge 2: analysis and results

    by P Hartley, A Bonaldi, R Braun, J N H S Aditya, S Aicardi, L Alegre, A Chakraborty, X Chen, S Choudhuri, A O Clarke, et al.

    The Square Kilometre Array Observatory (SKAO) will explore the radio sky to new depths in order to conduct transformational science. SKAO data products made available to astronomers will be correspondingly large and complex, requiring the application of advanced analysis techniques to extract key science findings. To this end, SKAO is conducting a series of Science Data Challenges, each designed to familiarize the scientific community with SKAO data and to drive the development of new analysis techniques. We present the results from Science Data Challenge 2 (SDC2), which invited participants to find and characterize 233 245 neutral hydrogen (H i) sources in a simulated data product representing a 2000 h SKA-Mid spectral line observation from redshifts 0.25–0.5. Through the generous support of eight international supercomputing facilities, participants were able to undertake the Challenge using dedicated computational resources. Alongside the main challenge, ‘reproducibility awards’ were made in recognition of those pipelines which demonstrated Open Science best practice. The Challenge saw over 100 participants develop a range of new and existing techniques, with results that highlight the strengths of multidisciplinary and collaborative effort. The winning strategy – which combined predictions from two independent machine learning techniques to yield a 20 per cent improvement in overall performance – underscores one of the main Challenge outcomes: that of method complementarity. It is likely that the combination of methods in a so-called ensemble approach will be key to exploiting very large astronomical data sets.


  • Open Access

    A structural homologue of the plant receptor D14 mediates responses to strigolactones in the fungal phytopathogen Cryphonectria parasitica

    by Valentina Fiorilli, Marco Forgia, Alexandre de Saint Germain, Giulia D’arrigo, David Cornu, Philippe Le Bris, Salim Al‐Babili, Francesca Cardinale, Cristina Prandi, Francesca Spyrakis, et al.

    Summary: Strigolactones (SLs) are plant hormones and important signalling molecules required to promote arbuscular mycorrhizal (AM) symbiosis. While in plants an α/β‐hydrolase, DWARF14 (D14), was shown to act as a receptor that binds and cleaves SLs, the fungal receptor for SLs is unknown. Since AM fungi are currently not genetically tractable, in this study, we used the fungal pathogen Cryphonectria parasitica, for which gene deletion protocols exist, as a model, as we have previously shown that it responds to SLs. By means of computational, biochemical and genetic analyses, we identified a D14 structural homologue, CpD14. Molecular homology modelling and docking support the prediction that CpD14 interacts with and hydrolyses SLs. The recombinant CpD14 protein shows α/β hydrolytic activity in vitro against the SLs synthetic analogue GR24; its enzymatic activity requires an intact Ser/His/Asp catalytic triad. CpD14 expression in the d14‐1 loss‐of‐function Arabidopsis thaliana line did not rescue the plant mutant phenotype. However, gene inactivation by knockout homologous recombination reduced fungal sensitivity to SLs. These results indicate that CpD14 is involved in SLs responses in C. parasitica and strengthen the role of SLs as multifunctional molecules acting in plant–microbe interactions.