Florent Leclercq

Invited Talks at International Meetings

• 13. Potentially brilliant: a history of boundaries
  02-07-2026, The Random Universe conference, Imperial College London, London, UK slides abstract
  The story of spatial COmoving Lagrangian Acceleration (sCOLA) is a story of boundaries, and of one well-timed suggestion that turned a stalled idea into a working method. sCOLA promised something remarkable: split a cosmological volume into independent tiles, evolve each one with no communication whatsoever, and recover a full simulation with perfect parallelism. The catch was the boundaries. Each tile is a small, non-periodic window onto a larger Universe, and getting the gravitational potential right at its edges proved decisive. Early attempts were simply not accurate enough, and the approach risked remaining a curiosity. The breakthrough came when Andrew suggested approximating the boundary potential with the Linearly Evolving Potential, which finally unlocked the accuracy the method needed, as published in 2020. From there, the idea has travelled: COCA (2024) generalises the comoving frame from perturbation theory to a machine-learnt one, while sCOLA lightcones (2026) evolve each tile only until it exits the observer's past lightcone. Sixty years in, Andrew is still keeping us all within bounds.
• 12. Implicit likelihood inference from galaxy survey data with robustness to model misspecification
  12-01-2024, Simulation based inference in Astrophysics, Royal Astronomical Society Specialist Discussion Meeting, London, UK slides abstract
  While the next generation of surveys soon starting will generate orders of magnitude more data than previously, it is becoming increasingly clear that traditional techniques are not up to the challenge of fully exploiting the raw data. I will present recent methodological advances, aiming at inferring the initial matter power spectrum and cosmological parameters from arbitrarily complex forward models of galaxy surveys where all relevant statistics can be determined from numerical simulations. The methods (SELFI and BOLFI) allow us to push analyses further into the non-linear regime than state-of-the-art backward modelling techniques. Importantly, they enable a check for model misspecification at the level of the initial matter power spectrum before final inference of cosmological parameters. I will present an application to a Euclid-like configuration.
• 11. Forward modelling the large-scale structure: field-level and implicit likelihood inference
  29-11-2022, Rubin LSST France meeting, Laboratoire de physique nucléaire et de hautes énergies, Paris, France slides abstract
  While the next generation of surveys soon starting will generate orders of magnitude more data than previously, it is becoming increasingly clear that traditional techniques are not up to the challenge of fully exploiting the raw data. The last few years have seen vast progress in the field of probabilistic large-scale structure inference, which differs in intent from traditional measurements of statistical summaries from galaxy survey catalogues. In this talk, I will review recent methodological advances. First, I will present the field-level likelihood-based approach, which allows the ab initio simultaneous analysis of the formation history and morphology of the cosmic web. Second, I will discuss the inference of physical parameters when the likelihood is implicitly defined by a "black-box" simulator.
• 10. The multi-stream local Universe
  28-01-2020, The Cosmic Web in the Local Universe, workshop at the Lorentz Center, Leiden, The Netherlands slides abstract
  Recent developments of Bayesian large-scale structure inference technology naturally bring in new ways of testing the standard paradigm of cosmic structure formation. I will present a cosmographic analysis of the dark matter distribution and its evolution, referred to as the dark matter phase-space sheet, in the nearby universe as traced by the 2M++ galaxy catalogue. With a few examples, I will argue that accessing the phase-space structure of dark matter in galaxy surveys opens the way for new confrontations of observational data and theoretical models.
• 9. Primordial power spectrum and cosmology from black-box galaxy surveys, Prospects for Euclid
  12-12-2019, CoSyne: Cosmological Synergies in the upcoming decade, conference at the Institut d'Astrophysique de Paris, Paris, France slides
• 8. Cosmic web analysis and information theory
  17-06-2019, The Cosmic Web: From Galaxies to Cosmology, workshop at the Higgs Centre, Royal Observatory, Edinburgh, UK slides abstract
  Recent developments of Bayesian large-scale structure inference technology naturally bring in a connection between cosmic web analysis and information theory. I will discuss the Shannon entropy of structure-type probability distributions, propose a decision criterion for classifying structures in the presence of uncertainty, and introduce utility functions for the optimal choice of a cosmic web classifier. As showcases, I will discuss the phase-space structure of nearby dark matter, the discrimination of dark energy models from the cosmic web and an approach inspired by supervised machine learning for predicting galaxy colours given their large-scale environment.
• 7. Bayesian inference with black-box cosmological models
  30-05-2019, Machine Learning in Astronomy, Royal Statistical Society workshop at Lancaster University, Lancaster, UK slides abstract
  Large-scale astronomical surveys carry opportunities for testing physical theories about the origin and evolution of the Universe. Advancing the research frontier requires solving challenging and unique statistical problems, to unlock the information content of massive and complex data streams. In this talk, I will present recent methodological advances, aiming at fitting cosmological data with "black-box" numerical models. I will discuss two different solutions, depending on the scenario: Bayesian optimisation and Taylor-expansion of the simulator.
• 6. Bayesian optimisation for likelihood-free cosmological inference
  22-10-2018, Conference "Methods for Statistical Inference" during the IHP Trimester 2018: Analytics, Inference, and Computation in Cosmology, Institut Henri Poincaré, Paris, France slides
• 5. Cosmic web analysis and Information Theory
  06-07-2018, The Information Universe symposium, Groningen, The Netherlands slides abstract
  Recent developments of Bayesian large-scale structure inference technology naturally bring in a connection between cosmic web analysis and information theory. I will discuss the Shannon entropy of structure-type probability distributions and the information gain due to Sloan Digital Sky Survey galaxies, propose a decision criterion for classifying structures in the presence of uncertainty, and introduce utility functions for the optimal choice of a cosmic web classifier, specific to the application of interest. As showcases, I will discuss the phase-space structure of nearby dark matter, the discrimination of dark energy models from the cosmic web and an approach inspired by supervised machine learning for predicting galaxy colours given their large-scale environment.
• 4. Bayesian large-scale structure inference
  08-07-2016, BAO & RSD: dark light on obscure acronyms, workshop at the Sexten-CfA, Sesto, Italy slides abstract
  The last few years have seen vast progress in the field of probabilistic large-scale structure inference, which differs in intent from traditional measurements of statistical summaries from galaxy survey catalogs. The resulting set of techniques aims at a complete statistical characterization and at a detailed cosmographic description of the large-scale structure.
  I will review recent progress in the field, discussing in particular, in order of increasing complexity, the Gaussian (Wiener filtering), log-normal, and physical structure formation approach as large-scale structure likelihoods. In doing so, I will present the required statistical tools and show respective data applications. I will finally discuss the opportunities and challenges associated to the inclusion of physical effects such as baryon acoustic oscillations and redshift-space distortions in the data model; particularly toward the goal of inferring cosmological parameters from survey data.
• 3. Cosmic web analysis and information theory: some recent results
  10-06-2016, Mapping the Cosmic Web: A Frontier in Cosmology and Astrophysics, workshop at the Royal Astronomical Society, London, UK slides abstract
  Recent developments of Bayesian large-scale structure inference technology naturally bring in a connection between cosmic web analysis and information theory. I will discuss the Shannon entropy of structure-type probability distributions and the information gain due to Sloan Digital Sky Survey galaxies, propose a decision criterion for classifying structures in the presence of uncertainty, and introduce utility functions for the optimal choice of a cosmic web classifier, specific to the application of interest. As showcases, I will discuss the discrimination of dark energy models from the cosmic web and an approach inspired by supervised machine learning for predicting galaxy colours given their large-scale environment.
• 2. Constrained simulations of the dynamic cosmic web
  27-11-2014, Deciphering the dynamic Universe, confronting predictions with observations, workshop at the Max-Planck-Institut für Astrophysik, Garching, Germany slides abstract
  Building upon an innovative statistical data analysis approach designed for the simultaneous analysis of the formation history and morphology of the large-scale structure of the inhomogeneous Universe, I will discuss the primordial and late-time cosmic web in the Sloan Digital Sky Survey DR7 volume.
  This project has led to the first quantitative reconstruction of the cosmological initial conditions from galaxies, an exceptionally detailed tidal classification of the cosmic web underlying the observed galaxy distribution, and a new, enhanced catalog of cosmic voids probed at the level of the dark matter distribution, deeper than with the galaxies. In the future, the approach is expected to yield insight into galaxy formation physics, the dynamic properties of a dark energy component, the effects of the inhomogeneous large-scale structure on photons and the statistical properties of the early Universe.
  As three-dimensional large-scale structure surveys contain a wealth of information that cannot be trivially extracted due to the non-linear dynamical evolution of the density field, I will discuss methods designed to improve upon previous techniques by including non-Gaussian and non-linear data models for the description of late-time structure formation.
• 1. Bayesian inference of dark matter voids in galaxy surveys
  21-08-2014, Cosmic Voids in the Next Generation of Galaxy Surveys, workshop at CCAPP, Columbus, Ohio, USA slides abstract
  I will describe an innovative statistical data analysis approach designed for the ab initio simultaneous analysis of the formation history and morphology of the large-scale structure of the inhomogeneous Universe. As a physical application, I will describe the construction of new, enhanced catalogs of cosmic voids probed at the level of the dark matter distribution, deeper than with the galaxies. This approach greatly alleviates the problems due to the sparsity and biasing of tracers for cosmological interpretation of final results.
  This framework is extremely general for application to the current and next generation of galaxy surveys. As an example, I will demonstrate its application to the Sloan Digital Sky Survey data release 7 and discuss the properties of dark matter voids in the Sloan volume.

Invited Seminars

• 18. Hopes and challenges in information science for cosmology
  03-05-2024, Seminar at the Centre de Recherche Astrophysique de Lyon (CRAL), Lyon, France slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. While the next generation of surveys soon starting will generate orders of magnitude more data than previously, it is becoming increasingly clear that traditional techniques are not up to the challenge of fully exploiting the raw data. The last decade has seen vast progress in the field of information science, with routine large-scale applications of machine learning. Many of these developments have been driven by (astro-)physics and by the evolution of computing hardware. However, the story started much earlier with an epistemological controversy: how do we analyse evidence and change our minds as we get new information? How do we make rational decisions in the face of uncertainty? I will review some of these historical aspects and discuss some past successes and future promises of information science for cosmology, in applications such as speeding up models, going beyond approximations, finding the information content, and dealing with complex inference tasks. Throughout the talk, I will present the outcomes of some recent methodological advances. I will then turn to the future and highlight some opportunities and challenges for the field.
• 17. Hopes and challenges in information science for cosmology
  21-10-2022, IAP colloquium, Institut d'Astrophysique de Paris, Paris, France slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. Answering such physical questions requires extracting information from large astronomical data sets with sufficient accuracy and precision. While the next generation of surveys soon starting will generate orders of magnitude more data than previously, it is becoming increasingly clear that traditional techniques are not up to the challenge of fully exploiting the raw data. With next-generation experiments, advancing the research frontier will require solving challenging and unique statistical problems, to unlock the information content of massive and complex data streams.
  The last decade has seen vast progress in the field of information science, with routine large-scale applications of machine learning. Many of these developments have been driven by (astro-)physics and by the evolution of computing hardware. However, the story started much earlier with an epistemological controversy: how do we analyse evidence and change our minds as we get new information? How do we make rational decisions in the face of uncertainty? I will review some of these historical aspects and discuss some past successes and future promises of information science for cosmology, in applications such as speeding up models, going beyond approximations, finding the information content, and dealing with complex inference tasks. Throughout the talk, I will present the outcomes of some recent methodological advances. I will then turn to the future and highlight some opportunities and challenges for the field.
• 16. High-performance computing and high-performance data analysis in cosmology
  06-07-2022, Maison de la Simulation, CEA, Saclay, France slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. Advancing the research frontier requires solving unique and challenging statistical and computational problems, to unlock the information content of massive and complex data streams. In this talk, I will present some recent methodological advances in high-performance computing and high-performance data analysis inspired by research in cosmology. I will then turn to the future and highlight some challenges and opportunities for the next decade in the field of cosmostatistics.
• 15. Bayesian analyses of galaxy surveys
  24-11-2021, Astrophysics Group, Imperial College London (remote seminar), London, UK slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. Advancing the research frontier requires solving unique and challenging statistical problems, to unlock the information content of massive and complex data streams. In this talk, I will present the outcomes of recent methodological advances: the estimation of physical parameters using simulation-based inference techniques, and the reconstruction of our cosmic history using field-based likelihoods. I will then turn to the future and highlight some challenges and opportunities for the next decades in the field of astrostatistics.
• 14. Bayesian analyses of galaxy surveys
  09-06-2021, School of Physics and Astronomy, Queen Mary University of London (remote seminar), London, UK slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. Advancing the research frontier requires solving challenging and unique statistical problems, to unlock the information content of massive and complex data streams. In this talk, I will present the outcomes of recent methodological advances: the reconstruction of our cosmic history, and the estimation of physical parameters using "black-box" simulators. I will then turn to the future and highlight some challenges and opportunities for the next generation of galaxy surveys.
• 13. Perfectly parallel cosmological simulations using spatial comoving Lagrangian acceleration
  06-04-2020, Institute for Advanced Study/Princeton University Joint Astrophysics Colloquium (remote seminar), Princeton, New Jersey, USA slides abstract
  I will introduce a new, perfectly parallel approach to simulate cosmic structure formation, based on the spatial COmoving Lagrangian Acceleration (sCOLA) framework. Building upon a hybrid analytical and numerical description of particles' trajectories, sCOLA allows an efficient tiling of a cosmological volume, where the dynamics within each tile is computed independently. I will show that cosmological simulations at the degree of accuracy required for the analysis of the next generation of surveys can be run in drastically reduced wall-clock times and with very low memory requirements, and discuss perspectives for computing future larger and higher-resolution cosmological simulations, taking advantage of a variety of hardware architectures.
• 12. Bayesian analyses of galaxy surveys
  11-11-2019, Cosmology lunch seminar, DAMTP, Institute of Astronomy & Cavendish Astrophysics, University of Cambridge, Cambridge, UK slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. Advancing the research frontier requires solving challenging and unique statistical problems, to unlock the information content of massive and complex data streams. In this talk, I will present the outcomes of recent methodological advances: the reconstruction of our cosmic history, and the estimation of physical parameters using "black-box" simulators. I will then turn to the future and highlight some challenges and opportunities for the next generation of galaxy surveys.
• 11. Bayesian analyses of galaxy surveys
  02-05-2019, Institut de Recherche en Astrophysique et Planétologie (IRAP), Toulouse, France slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. Advancing the research frontier requires solving challenging and unique statistical problems, to unlock the information content of massive and complex data streams. In this talk, I will present the outcomes of recent methodological advances: the reconstruction of our cosmic history, and the estimation of physical parameters using "black-box" simulators. I will then turn to the future and highlight some challenges and opportunities for the next generation of galaxy surveys.
• 10. Methods for Cosmic Web Analysis
  29-10-2018, Seminar during the IHP Trimester 2018: Analytics, Inference, and Computation in Cosmology, Institut Henri Poincaré, Paris, France slides
• 9. Bayesian large-scale structure inference, likelihood-based and likelihood-free approaches
  13-04-2018, Astrophysics seminar, University of Helsinki, Helsinki, Finland slides
• 8. Bayesian large-scale structure inference, likelihood-based and likelihood-free approaches
  31-01-2018, Astrophysics Group, Imperial College London, London, UK slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. The last few years have seen vast progress in the field of probabilistic large-scale structure inference, which differs in intent from traditional measurements of statistical summaries from galaxy survey catalogues. In this talk, I will present the outcomes of recent methodological advances.
  A likelihood-based approach allows the ab initio simultaneous analysis of the formation history and morphology of the cosmic web. To this end, the BORG (Bayesian Origin Reconstruction from Galaxies) algorithm infers the primordial density fluctuations and produces physical reconstructions of the dark matter distribution that underlies observed galaxies, by assimilating the survey data into a cosmological structure formation model. I will demonstrate the application of BORG to the Sloan Digital Sky Survey data and describe subsequent cosmic web analyses.
  I will also consider the opportunities and challenges associated to the likelihood-free inference of physical parameters from survey data. The proposed approximate Bayesian approach fits large-scale structure data with a very general numerical model: a noisy non-linear dynamical system with millions of variables. By combining a regression of the distance between simulated and observed data with Bayesian optimisation, it reduces the number of required simulations by several orders of magnitude.
• 7. Farewell talk
  16-05-2017, Institute of Cosmology and Gravitation, University of Portsmouth, Portsmouth, UK slides abstract
  The last decade has seen vast progress in the field of probabilistic large-scale structure inference, which differs in intent from traditional measurements of statistical summaries from galaxy survey catalogues. Before leaving the ICG in a few months, I will review my contributions of the last two years to the field, highlighting two different approaches: likelihood-based and likelihood-free.
  Building upon the application of the likelihood-based method BORG to Sloan Digital Sky Survey data, I will show detailed characterisations of dynamic cosmic web environments. In doing so, I will discuss a natural connection with information theory.
  I will then present ongoing work about likelihood-free inference with generative cosmological models. The proposed approach reduces the number of required simulations by several orders of magnitude, yet the computational cost remains a major challenge. As an answer, in the last part, I will discuss an innovative approach for embarrassingly parallel cosmological simulations, based on a spatial splitting of the considered volume.
• 6. How is the cosmic web woven? A Bayesian approach
  22-06-2016, Special Universe talk at the Excellence Cluster Universe, Garching, Germany slides abstract
  BORG (Bayesian Origin Reconstruction from Galaxies) is an inference engine that derives the initial conditions given a cosmological model and the survey data, and produces physical reconstructions of the underlying large-scale structure by assimilating the data into the model. Building upon the application of BORG to the Sloan Digital Sky Survey data, I will present detailed characterizations of dynamic cosmic web type environments.
  These developments naturally bring in a connection between cosmic web analysis and information theory. I will discuss the Shannon entropy of structure-type probability distributions and the information gain due to Sloan Digital Sky Survey galaxies, propose a decision criterion for classifying structures in the presence of uncertainty, and introduce utility functions for the optimal choice of a cosmic web classifier, specific to the application of interest. As showcases, I will discuss the discrimination of dark energy models from the cosmic web and an approach inspired by supervised machine learning for predicting galaxy colours given their large-scale environment.
• 5. Probabilistic large-scale structure inference, cosmic web analysis and information theory
  25-04-2016, University of Sussex, Brighton, UK slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. I will present an innovative statistical approach for the ab initio simultaneous analysis of the formation history and morphology of the cosmic web: the BORG (Bayesian Origin Reconstruction from Galaxies) algorithm infers the primordial density fluctuations and produces physical reconstructions of the dark matter distribution that underlies observed galaxies, by assimilating the survey data into a cosmological structure formation model.
  I will demonstrate the application of BORG to the Sloan Digital Sky Survey data and describe the formation history of the observed large-scale structure. As three-dimensional surveys contain a wealth of information at smaller scales, I will discuss opportunities and challenges linked to including non-linear data models for the description of structure formation.
  Building upon inference results, I will present a detailed characterization of dynamic cosmic web type environments. These developments naturally bring in a connection between cosmic web analysis and information theory. In particular, I will discuss the Shannon entropy of the structure-type probability distribution and the information gain due to SDSS galaxies. I will also propose a decision criterion for classifying structures in the presence of uncertainty, and introduce utility functions for the optimal choice of a cosmic web classifier, specific to the application of interest.
• 4. Bayesian large-scale structure inference and cosmic web analysis
  08-12-2015, Institute of Cosmology and Gravitation, University of Portsmouth, Portsmouth, UK slides
• 3. Bayesian large-scale structure inference and cosmic web analysis
  02-03-2015, GReCO seminar, Institut d'Astrophysique de Paris, Paris, France slides abstract
  Ideally, cosmological surveys should be analyzed in terms of the joint constraints they place on the initial conditions from which structure originates and on their subsequent gravitational evolution. In this talk, I will describe an innovative statistical data analysis approach designed for the ab initio simultaneous analysis of the formation history and morphology of the large-scale structure of the inhomogeneous Universe.
  As three-dimensional large-scale structure surveys contain a wealth of information that cannot be trivially extracted due to the non-linear dynamical evolution of the density field, I will discuss methods designed to improve upon previous techniques by including non-Gaussian and non-linear data models for the description of late-time structure formation.
  Through the talk, I will demonstrate the application of our inference framework to the Sloan Digital Sky Survey data release 7 and describe the primordial and late-time large-scale structure in the Sloan volume. I will show how the approach has led to the first quantitative reconstructions of the cosmological initial conditions from galaxies, an exceptionally detailed characterization of the dynamic cosmic web underlying the observed galaxy distribution, and a new, enhanced catalog of cosmic voids probed at the level of the dark matter distribution, deeper than with the galaxies.
• 2. How did structure appear in the Universe? A Bayesian approach
  09-12-2014, Physics Division Research Progress Meeting, Lawrence Berkeley National Laboratory, Berkeley, California, USA slides abstract
  Ideally, cosmological surveys should be analyzed in terms of the joint constraints they place on the initial conditions from which structure originates and on their subsequent gravitational evolution. In this talk, I will describe an innovative statistical data analysis approach designed for the ab initio simultaneous analysis of the formation history and morphology of the large-scale structure of the inhomogeneous Universe.
  As three-dimensional large-scale structure surveys contain a wealth of information that cannot be trivially extracted due to the non-linear dynamical evolution of the density field, I will discuss methods designed to improve upon previous techniques by including non-Gaussian and non-linear data models for the description of late-time structure formation.
  Through the talk, I will demonstrate the application of our inference framework to the Sloan Digital Sky Survey data release 7 and describe the primordial and late-time large-scale structure in the Sloan volume. I will show how the approach has led to the first quantitative reconstructions of the cosmological initial conditions from galaxies, an exceptionally detailed characterization of the dynamic cosmic web underlying the observed galaxy distribution, and a new, enhanced catalog of cosmic voids probed at the level of the dark matter distribution, deeper than with the galaxies. Finally, I will present research plans aimed at further physical investigations using cosmological reconstructions.
• 1. Bayesian chrono-cosmography
  03-09-2014, Institute for Advanced Studies, Princeton, New Jersey, USA slides

Contributed Talks

• 17. Counterfactual-informed adaptive MCMC with conditional normalising flows
  15-12-2025, MaxEnt 2025 conference, Auckland, Aotearoa New Zealand slides abstract
  Markov Chain Monte Carlo (MCMC) methods are particularly powerful as they can correct for their mistakes via the Metropolis–Hastings (MH) acceptance test, providing an asymptotic guarantee of convergence to the target distribution. A key ingredient is the proposal distribution, which determines sampling efficiency. However, theory offers no general prescription for designing such a proposal distribution. In high-dimensional settings, one often finds that any naïve proposal yields no accepted samples which could be used as training data. Moreover, if the data model is non-differentiable, gradients are also unavailable as training data.
  We address the automatic design of a proposal distribution in scenarios where naïve choices result in near-zero acceptance rates and gradients cannot be employed. Specifically, we consider models comprising a signal (an arbitrary, non-differentiable function of the parameters) and additive Gaussian noise. We present a geometric interpretation of the MH test in this context, showing that each test defines a hyperplane that partitions data space into acceptance and rejection regions. We demonstrate that a rejection in the MCMC of interest would correspond to an acceptance in an alternative chain if the true data vector were replaced by a suitably chosen alternative data vector, close in the Mahalanobis sense.
  Building on this insight, we introduce a novel MCMC algorithm that augments traditional MH sampling with “reasoning with counterfactuals.” By recording not only {accepted parameter, true data} pairs, but also {rejected parameter, alternative data} pairs that would have led to acceptance, we construct a replay buffer for training a conditional normalising flow. This normalising flow serves as an independence proposal alongside a vanilla random-walk proposal. The result is a general-purpose, adaptive MCMC method with a proposal distribution that self-improves by learning from both accepted and rejected moves. We evaluate the performance of our algorithm on challenging Bayesian inference tasks, including field-level inference in cosmological data analysis.
• 16. Accelerated forward modelling of dark matter dynamics: ML-safety and perfect parallelism
  22-09-2025, Putting the Cosmic Large-scale Structure on the Map: Theory Meets Numerics, workshop at ESI, Vienna, Austria slides abstract
  I will present some recent progress in accelerating N-body simulations of cold dark matter, using temporal COmoving Computer Acceleration (tCOCA, Bartlett, Chiarenza, Doeser & Leclercq 2024) and spatial COmoving Lagrangian Acceleration (sCOLA, Leclercq et al. 2020, Aubin, Leclercq & Lavaux, in prep.). I will then discuss some perspectives.
• 15. Accelerated forward modelling of dark matter dynamics: ML-safety and perfect parallelism
  14-05-2025, Big Data, Big Questions: The Future of Cosmological Surveys, workshop at MIAPbP, Garching, Germany slides
• 14. Don't trust neural networks? Me neither, but here's how I use them anyway
  21-05-2024, Statistical Challenges in 21st Century Cosmology, Chania, Crete, Greece slides abstract
  Interpretability and accuracy are pivotal challenges in the application of machine learning to cosmology. If machines find something humans don't understand, how can we check (and trust) the results? In this presentation, I contend that addressing this concern is not always obligatory, whether machine learning is used to build a posterior approximator or an emulator of an expensive model. I will elucidate this argument through two case studies where the use of neural networks is safe by construction. My first example will involve information maximising neural networks to autonomously define statistical summaries for implicit likelihood inference. My second example will involve neural networks to emulate a frame of reference (rather than the simulation output) in a COLA-like framework for N-body simulations of dark matter particles.
• 13. Additional probes and alternative techniques for galaxy clustering
  25-01-2024, Symposium Euclid France 2024, Institut de Physique des Deux Infinis, Lyon, France slides abstract
  I will give an overview of the work developed in the Galaxy Clustering: Additional Probes (GC:AP) work package that I co-lead with Cora Uhlemann. Our work focuses on complementary approaches to main Euclid GC science, including: additional summary statistics (linear point BAO, one-point pdf), alternative data analysis techniques (forward modelling, implicit inference, field-based inference) and cross-correlations with other surveys/probes (velocity field reconstruction with distance tracers, multi-messenger).
• 12. Dealing with systematic effects: the issue of robustness to model misspecification
  28-11-2023, Debating the potential of machine learning in astronomical surveys #2 conference, Institut d'Astrophysique de Paris, Paris, France slides abstract
  Model misspecification is a long-standing problem for Bayesian inference: when the model differs from the actual data-generating process, posteriors tend to be biased and/or overly concentrated. This issue is particularly critical for cosmological data analysis in the presence of systematic effects. I will briefly review state-of-the-art approaches based on an explicit field-level likelihood, which sample known foregrounds and automatically report unknown data contaminations. I will then present recent methodological advances in the implicit likelihood approach, with arbitrarily complex forward models of galaxy surveys where all relevant statistics can be determined from numerical simulations. The method (Simulator Expansion for Likelihood-Free Inference, SELFI) allows to push analyses further into the non-linear regime than state-of-the-art backward modelling techniques. Importantly, it allows a check for model misspecification at the level of the initial matter power spectrum before final inference of cosmological parameters. I will present an application to a Euclid-like configuration.
• 11. SELFI enhanced: robustness to model misspecification and Euclid forecast
  28-11-2022, Euclid France Theory and Likelihood workshop, Institut d'Astrophysique de Paris, Paris, France slides abstract
  I will present recent methodological advances developed in the Additional Probes work package of the Galaxy Clustering SWG, aiming at inferring the initial matter power spectrum and cosmological parameters from arbitrarily complex forward models of galaxy surveys where all relevant statistics can be determined from numerical simulations. The method (Simulator Expansion for Likelihood-Free Inference, SELFI) allows to push analyses further into the non-linear regime than state-of-the-art backward modelling techniques. Importantly, it allows a check for model misspecification at the level of the initial matter power spectrum before final inference of cosmological parameters. I will present an application to a Euclid-like configuration, and discuss our progresses and work plan.
• 10. Simulation-based inference of Bayesian hierarchical models while checking for model misspecification
  19-07-2022, 41st MaxEnt2022 Conference, Institut Henri Poincaré, Paris, France slides abstract proceedings
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. Advancing the research frontier requires solving unique and challenging statistical problems, to unlock the information content of massive and complex data vectors. Traditional techniques used to analyse galaxy surveys involve a significant compression of the raw data, made of three-dimensional fields, into physically interpretable summary statistics, such as correlation functions. Unfortunately, standard assumptions made to write down a likelihood for these summary statistics can cause significant biases, a problem that is alleviated with simulation-based (likelihood-free) inference (SBI) (Leclercq & Heavens 2021). For this reason, SBI techniques have recently gained a lot of attention in the field of cosmological data analysis. Model misspecification is a major challenge that any inference approach to cosmological data analysis has to face in order to obtain unbiased results in the presence of incomplete physical modelling and/or unknown systematic effects. As a response, this paper will present recent methodological advances to perform simulation-based inference of a general class of Bayesian hierarchical models (BHMs), while checking for model misspecification.
• 9. Likelihood-free inference from galaxy surveys, Prospects for Euclid
  27-06-2019, CosmoGold, The golden age of cosmology from Planck to Euclid, Institut d'Astrophysique de Paris, Paris, France slides abstract
  I will present recent methodological advances, aiming at inferring the primordial matter power spectrum and cosmological parameters from arbitrarily complex forward models of galaxy surveys where all relevant statistics can be determined from numerical simulations, i.e. black-boxes. The method allows to push analyses further into the non-linear regime than state-of-the-art backward modelling techniques. I will present an application to a Euclid-like configuration and discuss prospects for likelihood-free inference from Euclid data.
• 8. How is the cosmic web woven? Inference with generative cosmological models
  28-08-2017, COSMO17 conference, Paris, France slides abstract
  The last few years have seen vast progress in the field of probabilistic large-scale structure inference, which differs in intent from traditional measurements of statistical summaries from galaxy survey catalogues. In this talk, I will present the outcomes of recent methodological advances, aiming at fitting large-scale structure data with a very general numerical model: a noisy non-linear dynamical system with millions of hidden variables.
  Building upon the application of the method to Sloan Digital Sky Survey data, I will show detailed characterisations of dynamic cosmic web environments in the nearby Universe. In doing so, I will discuss a natural connection with information theory.
  I will also consider the opportunities and challenges associated to the inference of cosmological parameters from survey data. The proposed approach reduces the number of required simulations by several orders of magnitude, yet the computational cost remains a major challenge. As an answer, in the last part, I will introduce an innovative approach for embarrassingly parallel cosmological simulations, based on a spatial splitting of the considered volume.
• 7. Cosmic web analysis and information theory: some recent results
  26-05-2016, Statistical Challenges in 21st Century Cosmology, Chania, Crete, Greece slides abstract
  BORG (Bayesian Origin Reconstruction from Galaxies) is an inference engine that derives the initial conditions given a cosmological model and the survey data, and produces physical reconstructions of the underlying large-scale structure by assimilating the data into the model. Building upon the application of BORG to the Sloan Digital Sky Survey data, I will present detailed characterizations of dynamic cosmic web type environments.
  These developments naturally bring in a connection between cosmic web analysis and information theory. I will discuss the Shannon entropy of the structure-type probability distribution and the information gain due to SDSS galaxies, and propose a decision criterion for classifying structures in the presence of uncertainty. I will also introduce utility functions for the optimal choice of a cosmic web classifier, specific to the application of interest. As showcases, I will discuss the discrimination of dark energy models from the cosmic web and an approach inspired by supervised machine learning for predicting galaxy colours given their large-scale environment.
• 6. Cosmic web analysis and information theory: some recent results
  13-01-2016, Statistical sampling and non-sampling methods in cosmology, workshop at the University of California, Berkeley, California, USA slides abstract
  The BORG (Bayesian Origin Reconstruction from Galaxies) algorithm is an inference engine that derives the initial conditions given a cosmological model and the survey data, and produces physical reconstructions of the underlying large-scale structure by assimilating the data into the model. It explores the joint posterior distribution of all parameters involved via efficient Hamiltonian Monte Carlo sampling.
  I will demonstrate the application of BORG to real galaxy catalogs and describe the primordial and late-time large-scale structure in the volume covered by the Sloan Digital Sky Survey main galaxy sample. Building upon these results, I will present a detailed characterization of dynamic cosmic web type environments.
  These developments naturally bring in a connection between cosmic web analysis and information theory. In particular, I will discuss the Shannon entropy of the structure-type probability distribution and the information gain due to SDSS galaxies. I will also propose a decision criterion for classifying structures in the presence of uncertainty, and introduce utility functions for the optimal choice of a cosmic web classifier, specific to the application of interest.
• 5. Probabilistic cartography of the large-scale structure
  21-08-2015, Rencontres du Vietnam 2015: Cosmology 50 years after CMB discovery, Quy Nhon, Vietnam slides abstract proceedings
  Cosmological surveys are sensitive both to the initial conditions from which cosmic structure originates and their subsequent gravitational evolution, and therefore constrain both of them jointly. The BORG (Bayesian Origin Reconstruction from Galaxies) algorithm is an inference engine that derives the initial conditions given a cosmological model and the survey data, and produces physical reconstructions of the underlying large-scale structure by assimilating the data into the model.
  In this talk, I will present the results of using BORG for the ab initio simultaneous analysis of the formation history and morphology of the cosmic web. I will demonstrate the application of our inference framework to real galaxy catalogs and describe the primordial and late-time large-scale structure in the nearby Universe. I will show how the approach has led to the first quantitative reconstructions of the cosmological initial conditions from galaxies, and an exceptionally detailed characterization of the dynamic cosmic web underlying the observed galaxy distribution. Finally, I will present probabilistic maps of secondary effects that are expected in the cosmic microwave background, given observations of the nearby large-scale structure.
• 4. How is the Cosmic Web Woven? A Bayesian Approach
  14-05-2015, Advanced Workshop on Cosmological Structures from Reionization to Galaxies: Combining Efforts from Analytical and Numerical Methods, ICTP, Trieste, Italy slides abstract
  Ideally, cosmological surveys should be analyzed in terms of the joint constraints they place on the initial conditions from which structure originates and on their subsequent gravitational evolution. In this talk, I will describe recent results that build upon the BORG (Bayesian Origin Reconstruction from Galaxies) algorithm, an innovative statistical data analysis approach designed for the ab initio simultaneous analysis of the formation history and morphology of the large-scale structure of the inhomogeneous Universe.
  Through the talk, I will demonstrate the application of our inference framework to the Sloan Digital Sky Survey data release 7 and describe the primordial and late-time large-scale structure in the Sloan volume. I will show how the approach has led to the first quantitative reconstructions of the cosmological initial conditions from galaxies, an exceptionally detailed characterization of the dynamic cosmic web underlying the observed galaxy distribution, and a new, enhanced catalog of cosmic voids probed at the level of the dark matter distribution, deeper than with the galaxies.
  As three-dimensional large-scale structure surveys contain a wealth of information that cannot be trivially extracted due to the non-linear dynamical evolution of the density field, I will discuss methods designed to improve upon standard techniques by including non-Gaussian and non-linear data models for the description of late-time structure formation.
• 3. How did structure appear in the Universe? A Bayesian approach
  27-08-2014, COSMO 2014 conference, Chicago, Illinois, USA slides abstract
  Establishing a quantitative link between cosmological observations and theories describing the early Universe has the potential to further our knowledge of fundamental physics on a wide range of energy and distance scales. In this talk, I will describe an innovative statistical data analysis approach designed for the ab initio simultaneous analysis of the formation history and morphology of the large-scale structure of the inhomogeneous Universe.
  I will demonstrate its application to the Sloan Digital Sky Survey data release 7 and describe the primordial and late-time large-scale structure in the Sloan volume. This approach has led to the first quantitative reconstructions of the cosmological initial conditions from galaxies, an exceptionally detailed characterization of the dynamic cosmic web underlying the observed galaxy distribution, and a new, enhanced catalog of cosmic voids probed at the level of the dark matter distribution, deeper than with the galaxies.
  Finally, as three-dimensional large-scale structure surveys contain a wealth of information that cannot be trivially extracted due to the non-linear dynamical evolution of the density field, I will discuss methods designed to improve upon previous approaches by including non-Gaussian and non-linear data models for the description of late-time structure formation.
• 2. Bayesian inference of the initial conditions from large-scale structure surveys
  26-06-2014, The Zel'dovich Universe: Genesis and Growth of the Cosmic Web, IAU Symposium 308, Tallinn, Estonia slides abstract proceedings
  Analysis of three-dimensional cosmological surveys has the potential to answer outstanding questions on the early Universe, and therefore on the very high energy physics at play during inflation. In this talk, I will describe recently proposed statistical data analysis methods designed to study the primordial large-scale structure via physical inference of the initial conditions in a fully Bayesian framework, and demonstrate its application to the Sloan Digital Sky Survey data release 7.
  I will illustrate the use of these results to probe fundamental physics: correlating reconstructed initial conditions with various galaxy properties reveals interesting relations between galaxies and primordial features of initial density fields and tidal shear tensors. Reconstruction of the large-scale environment also allows accessing cosmic voids at the level of the dark matter distribution, deeper than with the galaxies.
  Finally, as three-dimensional large-scale structure surveys contain a wealth of information which cannot be trivially extracted due to the non-linear dynamical evolution of the density field, I will present a method for fast generation of mock density fields improving upon previous approaches by including the salient features of the non-linear regime.
• 1. Bayesian large-scale structure inference: initial conditions and the cosmic web
  25-05-2014, Statistical Challenges in 21st Cosmology, IAU Symposium 306, Lisbon, Portugal slides abstract proceedings
  Establishing a link between cosmological observations and theories describing the early Universe is important because it can further our knowledge of fundamental physics on a wide range of energy and distance scales. In this talk, I will describe recently proposed statistical data analysis methods designed to study the primordial large-scale structure via physical inference of the initial conditions in a fully Bayesian framework, and demonstrate its application to the Sloan Digital Sky Survey data release 7. Our algorithm explores the joint posterior distribution of all parameters involved via efficient Hamiltonian Markov Chain Monte Carlo sampling.
  Building upon these results, I will show how to generate a set of data-constrained reconstructions of the present large-scale dark matter distribution. As a physical illustration, we apply a void identification algorithm to them. In this fashion, we access voids defined by the inferred dark matter field, not by galaxies, greatly alleviating the bias problem. In addition, the use of full-scale physical density fields yields a drastic reduction of statistical uncertainty in void catalogs. This new catalog is an enhanced data set for cross-correlation with other cosmological surveys.
  Finally, as three-dimensional large-scale structure surveys contain a wealth of information which cannot be trivially extracted due to the non-linear dynamical evolution of the density field, I will discuss methods designed to improve upon previous approaches by including the salient features of the non-linear regime.

Posters

• 1. Bayesian inference of dark matter voids in galaxy surveys
  22-03-2014, Rencontres de Moriond Cosmology 2014, La Thuile, Italy slides abstract proceedings
  Analysis of three-dimensional cosmological surveys has the potential to answer outstanding questions on the early Universe, and therefore on the very high energy physics at play during inflation. In this talk, I will describe recently proposed statistical data analysis methods designed to study the primordial large-scale structure via physical inference of the initial conditions in a fully Bayesian framework, and demonstrate its application to the Sloan Digital Sky Survey data release 7.
  I will illustrate the use of these results to probe fundamental physics: correlating reconstructed initial conditions with various galaxy properties reveals interesting relations between galaxies and primordial features of initial density fields and tidal shear tensors.
  Finally, as three-dimensional large-scale structure surveys contain a wealth of information which cannot be trivially extracted due to the non-linear dynamical evolution of the density field, I will present a method for fast generation of mock density fields improving upon previous approaches by including the salient features of the non-linear regime.

Other Talks

• 48. Counterfactual-informed adaptive MCMC with conditional normalising flows
  11-06-2026, Journée du GdR IASIS “Bayesian inference for inverse problems” slides
• 47. Forward modelling and statistical inference for cosmological data analysis
  28-10-2025, Habilitation thesis defence, Institut d'Astrophysique de Paris, Paris, France slides
• 46. COmoving Computer Acceleration: N-body simulations in an emulated frame of reference
  28-10-2024, Colloque national action Dark Energy 2024, Institut Henri Poincaré, Paris, France slides abstract
  Interpretability and accuracy are pivotal challenges in the application of machine learning to cosmology. If machines find something humans don't understand, how can we check (and trust) the results? In this presentation, I contend that addressing this concern is not always obligatory, when machine learning is used to build an emulator of an expensive model. I will elucidate this argument through a case study where the use of neural networks is safe by construction. COmoving Computer Acceleration (COCA) is a hybrid framework interfacing ML with an N-body simulator. The correct physical equations of motion are solved in an emulated frame of reference, so that any emulation error is corrected by design. This approach corresponds to solving for the perturbation of particle trajectories around the machine-learnt solution, which is computationally cheaper than obtaining the full solution, yet is guaranteed to converge to the truth as one increases the number of force evaluations.
• 45. Implicit Likelihood Inference while efficiently checking for survey systematics
  01-02-2024, Euclid Galaxy Clustering Science Working Group meeting, Laboratoire d'Astrophysique de Marseille, France slides
• 44. Additional probes and alternative techniques for galaxy clustering
  31-01-2024, Euclid Galaxy Clustering Science Working Group meeting, Laboratoire d'Astrophysique de Marseille, France slides
• 43. Simulation-based inference pipelines for Euclid data
  21-06-2023, Euclid Consortium meeting 2023, Copenhagen, Danemark slides
• 42. Likelihood-free large-scale structure inference with robustness to model misspecification
  15-06-2023, Cosmology & Gravitation seminar, Oskar Klein Centre, Stockholm University, Stockholm, Sweden slides abstract
  While the next generation of surveys soon starting will generate orders of magnitude more data than previously, it is becoming increasingly clear that traditional techniques are not up to the challenge of fully exploiting the raw data. I will present recent methodological advances, aiming at inferring the initial matter power spectrum and cosmological parameters from arbitrarily complex forward models of galaxy surveys where all relevant statistics can be determined from numerical simulations. The method (Simulator Expansion for Likelihood-Free Inference, SELFI) allows to push analyses further into the non-linear regime than state-of-the-art backward modelling techniques. Importantly, it allows a check for model misspecification at the level of the initial matter power spectrum before final inference of cosmological parameters. I will present an application to a Euclid-like configuration.
• 41. Hopes and challenges in data science for cosmology
  15-12-2022, ASNUM2022 : Journées de l'Action Spécifique Numérique de l'INSU, Lyon, France slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. With next-generation data, avancing the research frontier will require solving challenging and unique statistical problems, to unlock the information content of massive and complex data streams. It is therefore very timely to survey the landscape of data science and machine learning techniques, and to critically evaluate their usefulness for solving cosmological problems. In this talk, I will discuss some past successes and future promises, in applications such as speeding up models, going beyond approximations, finding the information content, and dealing with complex inference tasks. In doing so, I will present the outcomes of some recent methodological advances. I will then turn to the future and highlight some opportunities and challenges for the field.
• 40. Perfectly parallel cosmological simulations using spatial comoving Lagrangian acceleration
  17-11-2022, Colloque national action Dark Energy 2022, Marseille, France slides abstract
  I will discuss perspectives for building accelerated forward data models of galaxy surveys. In particular, I will introduce a perfectly parallel approach to simulate cosmic structure formation, based on the spatial COmoving Lagrangian Acceleration (sCOLA) framework. Building upon a hybrid analytical and numerical description of particles' trajectories, sCOLA allows an efficient tiling of a cosmological volume, where the dynamics within each tile is computed independently. I will show that cosmological simulations at the degree of accuracy required for the analysis of the next generation of surveys can be run in drastically reduced wall-clock times and with very low memory requirements, and discuss perspectives for computing future larger and higher-resolution cosmological simulations, taking advantage of a variety of hardware architectures.
• 39. Simulation-based inference, Bayesian hierarchical models, and model misspecification
  15-09-2022, Simons Collaboration on Learning the Universe, Annual Meeting 2022 (remote presentation), New York, USA slides
• 38. Thème : Grandes Structures \& Projet : Euclid
  16-05-2022, Séminaire interne de l'IAP, Jardins de l'Anjou, Mauges-sur-Loire, France slides
• 37. Forward modelling the large-scale structure: perfectly parallel simulations and simulation-based inference
  20-01-2022, London Cosmology Discussion Meeting (remote presentation), Royal Astronomical Society, London, UK slides
• 36. Forward modelling the large-scale structure: perfectly parallel simulations and simulation-based inference
  12-10-2020, Colloque Action Dark Energy (remote presentation), France slides abstract
  I will first introduce a new, perfectly parallel approach to simulate cosmic structure formation, based on the spatial COmoving Lagrangian Acceleration (sCOLA) framework. Building upon a hybrid analytical and numerical description of particles' trajectories, sCOLA allows an efficient tiling of a cosmological volume, where the dynamics within each tile is computed independently. I will show that cosmological simulations at the degree of accuracy required for the analysis of the next generation of surveys can be run in drastically reduced wall-clock times and with very low memory requirements.
  In a second part, I will discuss how such simulations can be used as "black-box" models within data analysis. I will focus on two recent algorithms (SELFI and BOLFI), aiming at inferring the primordial matter power spectrum and cosmological parameters. I will present an application to a Euclid-like configuration and discuss prospects for simulation-based inference from Euclid data.
• 35. Galaxy Clustering with Likelihood-Free Inference (GCLFI project): Prospects and Forecasts for Euclid
  20-04-2020, Euclid Galaxy Clustering Science Working Group (remote presentation) slides
• 34. Perfectly parallel cosmological simulations using spatial comoving Lagrangian acceleration
  07-04-2020, "Journal-club Univers" at the Institut d'Astrophysique de Paris (remote presentation), Paris, France slides
• 33. Simulator Expansion for Likelihood-Free Inference, Prospects for Euclid
  04-02-2020, Euclid (GC-WL-CG-LE3) meeting, Institut d'Astrophysique de Paris, Paris, France slides abstract
  I will present recent methodological advances, aiming at inferring the primordial matter power spectrum and cosmological parameters from arbitrarily complex forward models of galaxy surveys where all relevant statistics can be determined from numerical simulations, i.e. black-boxes. The method allows to push analyses further into the non-linear regime than state-of-the-art backward modelling techniques. I will present an application to a Euclid-like configuration and discuss prospects for likelihood-free inference from Euclid data.
• 32. Simulator Expansion for Likelihood-Free Inference, Prospects for Euclid
  17-12-2019, Euclid UK 2019 meeting, Royal Astronomical Society, London, UK slides abstract
  I will present recent methodological advances, aiming at inferring the primordial matter power spectrum and cosmological parameters from arbitrarily complex forward models of galaxy surveys where all relevant statistics can be determined from numerical simulations, i.e. black-boxes. The method allows to push analyses further into the non-linear regime than state-of-the-art backward modelling techniques. I will present an application to a Euclid-like configuration and discuss prospects for likelihood-free inference from Euclid data.
• 31. Bayesian inference with black-box cosmological models
  20-09-2019, Approximate Bayesian Computation and novel Bayesian approaches in cosmostatistics, workshop at the Université Clermont-Auvergne, Clermont-Ferrand, France slides abstract
  Large-scale astronomical surveys carry opportunities for testing physical theories about the origin and evolution of the Universe. Advancing the research frontier requires solving challenging and unique statistical problems, to unlock the information content of massive and complex data streams. In this talk, I will present recent methodological advances, aiming at fitting cosmological data with "black-box" numerical models. I will discuss two different solutions, depending on the scenario: Bayesian optimisation (BOLFI) and Taylor-expansion of the simulator (SELFI).
• 30. The Cosmic Web: Theory, Simulations, Observations, Reconstructions
  31-07-2019, Review talk, Astrophysics Group, Imperial College London, London, UK slides
• 29. Bayesian analyses of galaxy surveys
  08-07-2019, Cosmology group, Laboratoire Astroparticule & Cosmologie (APC), Paris, France slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. Advancing the research frontier requires solving challenging and unique statistical problems, to unlock the information content of massive and complex data streams. In this talk, I will present the outcomes of recent methodological advances: the reconstruction of our cosmic history, and the estimation of physical parameters using "black-box" simulators. I will then turn to the future and highlight some challenges and opportunities for the next generation of galaxy surveys.
• 28. Density reconstruction via Bayesian large-scale structure inference
  05-06-2019, Euclid Consortium meeting 2019, Helsinki, Finland slides
• 27. Likelihood-free inference techniques for cosmology
  22-03-2019, The Most Ancient Heavens, conference at the Royal Astronomical Society, London, UK slides
• 26. Density reconstruction via Bayesian large-scale structure inference
  17-12-2018, Euclid UK meeting 2018, University of Oxford, Oxford, UK slides abstract
  Several algorithms and software packages aiming at a Bayesian probabilistic analysis of the full three-dimensional matter field inferred from the galaxy distribution have recently met some success (see for example https://aquila-consortium.org). They differ in intent from traditional measurements of statistical summaries from galaxy catalogues.
  Within the "Additional probes" work package of the Galaxy Clustering SWG, we are extending these algorithms for Euclid-specific aspects (large volume, photometric redshifts), and aim ultimately at a joint inference of maps and cosmology from Euclid data. I will present Bayesian algorithms for density reconstruction, and discuss our progresses and work plan.
• 25. Inference with generative cosmological models
  20-06-2018, Astrophysics Group, University College London, London, UK slides abstract
  The last few years have seen vast progress in the field of probabilistic large-scale structure inference, which differs in intent from traditional measurements of statistical summaries from survey data. In this talk, I will present recent methodological advances, aiming at fitting cosmological data with a very general numerical model: a noisy non-linear dynamical system with an unrestricted number of latent variables. The main challenge is then the intractability of the likelihood, and therefore, the computational cost. The proposed strategy combines probabilistic modelling of the discrepancy between simulated and observed data with optimisation to facilitate likelihood-free inference. As a consequence, the number of required simulations is reduced by several orders of magnitude.
• 24. Simulation-based large-scale structure inference
  23-01-2018, "Journal-club Univers" at the Institut d'Astrophysique de Paris, Paris, France slides
• 23. Bayesian optimisation for likelihood-free cosmological inference
  29-03-2017, Theory group meeting, Institute of Cosmology and Gravitation, University of Portsmouth, Portsmouth, UK slides abstract
  I will present ongoing work about inference with generative cosmological models. I assume that only a small number of parameters are of interest, but that the process generating the data is very general: a noisy non-linear dynamical system with millions of hidden variables.
  The main challenge is then the intractability of the likelihood, and therefore, the computational cost. The proposed strategy combines probabilistic modelling of the discrepancy between simulated and observed data with optimisation to facilitate likelihood-free inference. As a consequence, the number of required simulations is reduced by several orders of magnitude.
  I will discuss prospects to reanalyse existing large-scale structure data sets, including thorough forward-modelling of all relevant physical and observational effects.
• 22. Cosmic web analysis in the SDSS main galaxy sample and implications for galaxy colors
  26-01-2017, "Journal-club Galaxies" at the Institut d'Astrophysique de Paris, Paris, France slides
• 21. The phase-space structure of nearby dark matter
  24-01-2017, "Journal-club Univers" at the Institut d'Astrophysique de Paris, Paris, France slides
• 20. The phase-space structure of nearby dark matter as constrained by the SDSS main galaxy sample
  15-12-2016, ICAP meeting, Institut d'Astrophysique de Paris, Paris, France slides
• 19. Cosmic web analysis with the BORG SDSS run
  31-10-2016, Bayesian large-scale structure workshop, Max-Planck-Institut für Astrophysik, Garching, Germany slides
• 18. Cosmic web analysis and information theory: some recent results
  05-01-2016, "Journal-club Univers" at the Institut d'Astrophysique de Paris, Paris, France slides
• 17. Bayesian large-scale structure inference and cosmic web analysis
  24-09-2015, PhD defence, Institut d'Astrophysique de Paris, Paris, France slides abstract
  Surveys of the cosmic large-scale structure carry opportunities for building and testing cosmological theories about the origin and evolution of the Universe. This endeavor requires appropriate data assimilation tools, for establishing the contact between survey catalogs and models of structure formation.
  In this thesis, we present an innovative statistical approach for the ab initio simultaneous analysis of the formation history and morphology of the cosmic web: the borg algorithm infers the primordial density fluctuations and produces physical reconstructions of the dark matter distribution that underlies observed galaxies, by assimilating the survey data into a cosmological structure formation model. The method, based on Bayesian probability theory, provides accurate means of uncertainty quantification.
  We demonstrate the application of borg to the Sloan Digital Sky Survey data and describe the primordial and late-time large-scale structure in the observed volume. We show how the approach has led to the first quantitative inference of the cosmological initial conditions and of the formation history of the observed structures.
  We then use these results for several cosmographic projects aiming at analyzing and classifying the large-scale structure. In particular, we build an enhanced catalog of cosmic voids probed at the level of the dark matter distribution, deeper than with the galaxies. We present detailed probabilistic maps of the dynamic cosmic web, and offer a general solution to the problem of classifying structures in the presence of uncertainty.
  The results described in this thesis constitute accurate chrono-cosmography of the inhomogeneous cosmic structure.
• 16. Bayesian large-scale structure inference and cosmic web analysis
  30-04-2015, Large-scale structure group meeting, Institute of Cosmology and Gravitation, University of Portsmouth, Portsmouth, UK slides
• 15. Bayesian large-scale structure inference and cosmic web analysis
  03-04-2015, Journée des thèses, Institut d'Astrophysique de Paris, Paris, France slides abstract
  Ideally, cosmological surveys should be analyzed in terms of the joint constraints they place on the initial conditions from which structure originates and on their subsequent gravitational evolution.
  In this talk, I will describe progress towards simultaneous analysis of the formation history and web-like morphology of the large-scale structure of the inhomogeneous Universe. I will then discuss various projects aiming at segmenting the cosmic web into different structure types (voids, sheets, filaments and clusters).
• 14. Bayesian large-scale structure inference and cosmic web analysis
  05-03-2015, ICAP meeting, Institut d'Astrophysique de Paris, Paris, France slides
• 13. Cosmic order and complexity
  25-02-2015, YMCA, Institut d'Astrophysique de Paris, Paris, France slides abstract
  What banged at the Big Bang? How did complex structures take shape? Ideally, cosmological surveys should be analyzed in terms of the joint constraints they place on the initial conditions from which structure originates and on their subsequent gravitational evolution.
  In this talk, I will describe progress towards simultaneous analysis of the formation history and web-like morphology of the large-scale structure of the inhomogeneous Universe. Crucial for this aim is developing efficient tools for assimilating data into the forecasts of a physical model of cosmic structure formation and quantifying uncertainty within a fully probabilistic approach.
  I will keep the discussion at introductory level and focus on cosmographic results. For a more advanced discussion and comprehensive description of the methodology, I invite you to attend my GReCO seminar on March 2.
• 12. Cosmic order and complexity
  12-12-2014, Elbereth 2014 conference, Institut d'Astrophysique de Paris, Paris, France slides abstract
  What banged at the Big Bang? How did complex structures take shape? Ideally, cosmological surveys should be analyzed in terms of the joint constraints they place on the initial conditions from which structure originates and on their subsequent gravitational evolution.
  In this talk, I will describe progress towards simultaneous analysis of the formation history and morphology of the large-scale structure of the inhomogeneous Universe. Crucial for this aim is developing efficient tools for assimilating data into the forecasts of a physical model of cosmic structure formation and quantifying uncertainty within a fully probabilistic approach.
• 11. Bayesian inference and cosmic web classification
  15-05-2014, ICAP meeting, Institut d'Astrophysique de Paris, Paris, France slides
• 10. Bayesian large-scale structure inference: initial conditions and cosmic voids
  09-04-2014, YMCA, Institut d'Astrophysique de Paris, Paris, France slides abstract
  Establishing a link between cosmological observations and theories describing the early Universe is important because it can further our knowledge of fundamental physics on a wide range of energy and distance scales. In this talk, I will review statistical methods designed to learn about primordial physics by connecting information from cosmological probes to the statistical properties of the initial conditions of the Universe. I will describe recently proposed statistical data analysis methods designed to study the primordial large-scale structure via physical inference of the initial conditions in a fully Bayesian framework, and demonstrate its application to the Sloan Digital Sky Survey data release 7.
  Building upon these results, I will show how to generate a set of data-constrained reconstructions of the present large-scale dark matter distribution. As a physical illustration, we apply a void identification algorithm to them. In this fashion, we access voids defined by the inferred dark matter field, not by galaxies, greatly alleviating the bias problem. In addition, the use of full-scale physical density fields yields a drastic reduction of statistical uncertainty in void catalogs. This new catalog is an enhanced data set for cross-correlation with other cosmological surveys.
• 9. Bayesian inference of the initial conditions from large-scale structure surveys
  14-03-2014, Journée des thèses, Institut d'Astrophysique de Paris, Paris, France slides
• 8. Cosmostatistics: the initial conditions and the large-scale structure of the Universe
  27-11-2013, Elbereth 2013 conference, Institut d'Astrophysique de Paris, Paris, France slides
• 7. Cosmostatistics: the initial conditions and the large-scale structure of the Universe
  20-11-2013, PhD, Astroparticules & Cosmologie, Paris, France slides abstract
  At present, observations of the three-dimensional large-scale structure is one of the major sources of information on the origin of the Universe, and therefore on the very high energy physics at play during inflation. In this talk, I will discuss methods designed to learn about primordial physics by connecting information from the large-scale structure of the Universe to the statistical properties of its initial conditions. I will describe recently proposed methods for the physical inference of the initial conditions in a Bayesian framework, and application of these methods to the data of the SDSS DR7. Three-dimensional large-scale structure surveys contain a wealth of information which cannot be trivially extracted due to the non-linear dynamical evolution of the density field. In this context, modern cosmological data analysis has an increasing demand for numerically efficient models of structure formation. I will discuss perspectives to extend the tools used for the reconstruction of the initial conditions.
• 6. Cosmostatistics: the initial conditions and the large-scale structure of the Universe
  09-11-2013, Strings, Cosmology and Gravity Student Conference 2013, Max-Planck-Institute for Physics, Munich, Germany slides abstract
  At present, observations of the three-dimensional large-scale structure is one of the major sources of information on the origin of the Universe, and therefore on the very high energy physics at play during inflation. In this talk, I will discuss methods designed to learn about primordial physics by connecting information from the large-scale structure of the Universe to the statistical properties of its initial conditions. I will describe recently proposed methods for the physical inference of the initial conditions in a Bayesian framework. Three-dimensional large-scale structure surveys contain a wealth of information which cannot be trivially extracted due to the non-linear dynamical evolution of the density field. In this context, modern cosmological data analysis has an increasing demand for numerically efficient models of structure formation. I will discuss perspectives to extend the tools used for the reconstruction of the initial conditions.
• 5. Cosmostatistics: the initial conditions and the large-scale structure of the Universe
  09-07-2013, Post-Planck Cosmology Summer School, Les Houches, France slides
• 4. The initial conditions and the large-scale structure of the Universe
  12-06-2013, YMCA, Institut d'Astrophysique de Paris, Paris, France slides abstract
  At present, observations of the three-dimensional large-scale structure is one of the major sources of information on the origin of the Universe, and therefore on the very high energy physics at play during inflation. In this talk, I will discuss methods designed to learn about primordial physics by connecting information from the large-scale structure of the Universe to the statistical properties of its initial conditions.
  I will briefly review the standard methods used to deal with large surveys and the challenges we face.
  Due to the non-linearity of the gravitational processes involved in structure formation, modern cosmological data analysis has an increasing demand for numerically efficient statistical tools. I will describe recently proposed methods for the physical inference of the initial conditions in a Bayesian framework, and perspective to extend these tools.
  As many theoretical models involve non-trivial multi-point correlation functions, I will introduce some results regarding the influence of primordial non-Gaussianity in the large-scale structure and how these features can be used as probes of the inflationary era.
• 3. Optimizing Lagrangian perturbation theory in the mildly non-linear regime of cosmic structure formation via one-point remapping
  26-04-2013, Institute for Advanced Studies, Princeton, New Jersey, USA slides
• 2. The large-scale structure of the Universe as a probe of primordial physics
  29-03-2013, Journée des thèses, Institut d'Astrophysique de Paris, Paris, France slides
• 1. Large-scale structure and cosmic voids as probes of primordial physics
  12-12-2012, Elbereth 2012 conference, Institut d'Astrophysique de Paris, Paris, France slides abstract
  At present, observations of the three-dimensional large-scale structure is one of the major sources of information on the origin and evolution of the Universe. The detailed appearance of the presently observed matter distribution contains a record on its formation history.
  In this talk, we describe several methods designed to learn about primordial physics by connecting information from the large-scale structure of the Universe to the statistical properties of its initial conditions.
  Due to the non-linearity of the gravitational processes involved, at present we have just limited analytic understanding of structure formation in terms of perturbative expansions in Eulerian or Lagrangian representation. We describe a method designed to improve the correspondence between Lagrangian perturbation theory and full numerical simulations of gravitational large-scale structure formation, based on a remapping of the approximately evolved particle distribution using information extracted from N-body simulations.
  In order to mitigate the effect of gravitational non-linearity in galaxy clusters, we propose a new approach based on underdense regions such as cosmic voids in order to extract information from large-scale structure surveys. We explore the possibility of using a new estimator, the void-void two-point correlation function, and we argue that this quantity can be modeled easily up to redshift zero using Lagrangian perturbation theory.