&
dans le cadre de l'initiative d'excellence
A partir de Janvier 2022, découvrez mensuellement les thématiques de recherche des membres de l'ITI IRMIA++.
Les sessions ont habituellement lieu le 3e jeudi de chaque mois à 9h (sauf pendant les vacances universitaires).
Elles sont accessibles en présentiel à l'Université (en salle de conférence de l'IRMA, à l'ObAS ou à ICUBE)et les enregistrements des sessions passées sont disponibles sur la plateforme Pod.
9:00
IRMA Conference Room
Abstract: When an object is photographed, the resulting image is pixelated. The position of a point in such an image is described by integer coordinates, unlike that of a point on the original object, which is described by real coordinates. This transition from the usual Euclidean geometry describing the original object to the discrete geometry describing the obtained image, called digitization, causes significant information loss. If the resolution of the discrete image is too low compared to the level of detail of the original object, topological information and geometric quantities can be lost. It then becomes necessary to impose certain assumptions on this real object to allow the reconstruction of this information.
By modeling the digitization process, it is possible to ensure the reconstruction of the topology and geometric quantities of objects that meet certain assumptions. However, currently in digital geometry, the assumptions on real objects that guarantee the reconstruction of all this information are quite restrictive and do not allow the simultaneous inclusion of shapes whose boundary is a smooth curve and those whose boundary is a polygon.
To simultaneously address these two families of shapes, we propose a new assumption based on the concept of total curvature introduced by Milnor in 1950. This consists of locally limiting this total curvature on the boundary of the real object. This assumption, which includes shapes with smooth or polygonal boundaries, guarantees the reconstruction of topology and allows for bounding the errors of discrete estimators of geometric quantities.
About the speaker: Étienne Le Quentrec is assistant professor (maître de conférences) at ICube, member of IMAGeS team since September 2022. He defended his PhD thesis in 2021 on digital Geometry at ICube. He was also student at UFR in mathematics where he obtained his "agrégation" in 2016. His main research interests are digital topology and discrete estimation.
9:00
IRMA Conference room
Abstract: TBA
About the speaker: TBA
Abstract:
Since its inception in 2008 as the underlying technology behind Bitcoin, blockchain has evolved far beyond its cryptocurrency origins. Over the past decade, various blockchain systems have emerged, demonstrating diverse applications across industries. This talk provides a quick dive into blockchain analysis. I will cover the essentials of blockchain technology with a focus on consensus protocols like proof-of-work and proof-of-stake before discussing some applications in Finance and Insurance.
About the speaker: Pierre-Olivier Goffard is an Associate Professor at the University of Strasbourg, France. He conducts research at the intersection of applied probability, statistics, and their applications in finance and insurance. His areas of expertise encompass blockchain technology and Bayesian statistics. Pierre-Olivier actively participates in the actuarial science program at the University of Strasbourg, where he teaches courses on stochastic calculus and survival analysis.
Abstract: Numerous mathematical models of real-life processes pose major difficulties when it comes to their numerical simulation. Indeed, their dimensions are usually large, and their numerical resolution is often complex, requiring a great deal of computing power. The Vlasov-Poisson equation is one of such steep cost models.
Model order reduction (MOR) aims to build so-called reduced models which are valid approximations of their original counterparts with a reduced associated state space dimension. Doing so results in a reduced model that is much faster to solve, but at the cost of a certain accuracy.
Moreover, when the original model is Hamiltonian, it is critical to preserve this property at the reduced level. In this talk, I will show how to build Hamiltonian reduced models of the Vlasov-Poisson equation discretized with a Particle In Cell (PIC) method using classical tools from model order reduction. Then, I will introduce the benefits of adding neural networks to the process.
About the speaker: Guillaume Steimer is a Ph.D. student in applied mathematics and a member of the MACARON team at IRMA Strasbourg. He studies Hamiltonian systems and the construction of associated reduced models, specifically using machine learning techniques.
Abstract: Our galaxy, the Milky Way, can be modeled at zeroth order as an axisymmetric system at equilibrium, obeying the fundamental equations of galactic dynamics (the Vlasov-Poisson system of equations). Devising a precise non-axisymmetric model is, on the other hand, far from trivial. Using the conservation of the distribution function in infinitesimal phase- space patches following the Hamiltonian flow allows one to compute the current distribution function by integrating orbits backward in time to an axisymmetric equilibrium state. In this talk, I will show how we explored the vast parameter space of the bar and spiral arms with this method to establish the current most realistic dynamical non-axisymmetric model for the Milky Way disk.
About the speaker: Yassin Rany Khalil, after completing his studies at Ecole Polytechnique in 2021, has started an ITI-funded IRMIA++ PhD thesis at ObAS in 2022. His thesis deals with the detailed modelling of the stellar disk of our Milky Way galaxy.
Abstract: In this presentation, I will explain what vectorization is and how it is used by modern CPUs to increase peak performance. I will then discuss what it entails to vectorize code, focusing on the implementation of an efficient sorting algorithm, and why compilers often fail to do this automatically. Finally, I will present Autovesk, our tool for automatic vectorization. I will describe our solution for transforming a graph of scalar instructions into a graph of vectorized instructions.
About the speaker: Bérenger Bramas is a researcher (Chargé de Recherche) at Inria Nancy since October 2018. He is also a member of the ICube laboratory. He defended his PhD thesis in 2016 on the parallelization and optimization of the time-domain boundary element method for the wave equation. Subsequently, he worked as an HPC Expert at the Max Planck Supercomputing Center (MPCDF). His research interests focus on scientific computing, runtime systems, scheduling, software engineering for HPC, and automatic optimization/parallelization.
Presentation content:Click here to see the complete presentation
Abstract: Since 12 November 2019, the campus of the University of Strasbourg has been shaken several times by local earthquakes that have been linked to a deep geothermal energy project located in Vendenheim, 10km north of Strasbourg. The fear linked to these earthquakes is strongly questioning the value of using a local renewable resource underground as part of the energy transition. ITI GeoT has been heavily involved in characterising these seismic events, understanding their origin, questioning the technology and proposing new perspectives for better management of such projects. The seminar will illustrate the main observations and lessons from this recent technological history, focusing on the challenges posed by deep geothermal technologies, the development of subsurface observation techniques, including citizen science approaches, and the numerical modelling of subsurface physical responses through the development of digital twins.
About the speaker: Jean Schmittbuhl is CNRS researcher in geophysics at EOST/ITES – Ecole et Observatoire des Sciences de la Terre/Institut Terre et Environnement de Strasbourg). After multidisciplinary training in Earth Sciences at Ecole Normale Supérieure (ENS) Saint-Cloud/Lyon, he obtained an ‘agrégation’ in Natural Sciences in 1989 and a master's degree in Physics of Liquids at the Pierre-et-Marie-Curie University in 1991. He defended his PhD thesis in Physics in 1994 at ENS Paris. He became a junior CNRS researcher in 1995 at the Geology laboratory of ENS Paris, after a post-doctoral visit at the Physics department of the University of Oslo. In 2004, he joined the ‘Institut de Physique du Globe de Strasbourg’ as a CNRS research director and became head of the experimental geophysics team. In 2012, he initiated and coordinated the LabEx G-eau-thermie Profonde, as well as a deep geothermal energy consortium involving Electricité de Strasbourg. Since 2021, he has been the director of the Interdisciplinary Thematic Institute Geosciences for the Energy Transition (ITI GeoT) at the University of Strasbourg. He was a member of the expert committee for the industrial accident of the Vendenheim deep geothermal project. His interdisciplinary research combines seismology, geomechanics, physics of heterogeneous media, petrophysics, and geochemistry. He has developed innovative laboratory experiments, multiscale numerical modelling and geological and geophysical field observations. His research activity is focusing on geometry of fractures and faults, their frictional properties, fluid flow in fractured and porous massifs, thermo-hydro-chemo-mechanical couplings, and wave propagation in fractured media. His research applications include earthquake initiation processes, mechanisms of induced seismicity, imaging of deep reservoirs from microseismicity and ambient seismic noise, and deep geothermal projects.
Presentation content:Click here to see the complete presentation - Click here to download the content
This meeting will focus on interactions with biology. Our guests will be three colleagues from ITI IMCBio+ (Integrative Molecular & Cellular Biology). Ivan Tarassov, director, will give a brief a presentation of the themes covered by the ITI IMCBio+, followed by two short research talks by Joseph Schacherer and Nacho Molina.
Presentation content:Click here to see the complete presentation
About the speaker: Ivan Tarassov obtained his Master degree in Biochemistry & Molecular Biology in 1986 and PhD in 1990, both in Moscow State University. In 1992, he moved as a Postdoc (FEBS & EMBO fellowships) in Strasbourg, in IBMC. In 1996, he was recruited in the CNRS as CR1 and founded his own team in the GMGM unit. Between 2013 and 2023 he was the Director of the GMGM unit. In 2011, he founded the MitoCross labex and in 2022-2024 he is the coordinator of the ITI IMCBio+. His main scientific interests are mitochondrial functions and dysfunctions and mitochondrial diseases.
Presentation content:Click here to download the content of this talk
Abstract: Biology is currently undergoing an incredible revolution, enabled by the emergence of single-cell genomics. This advancement allows for the characterization of all cell types in the human body, leading to a systematic understanding of collective cell function in health and disease. However, computational biology faces the challenge of extracting valuable information and generating reliable predictions from this wealth of data. While deep learning methods have proven to be powerful tools for clustering and denoising data, their black-box nature limits interpretability and prediction power.
To address this limitation, we present an innovative approach that combines an interpretable variational autoencoder with biophysical modeling to characterize gene regulation using single-cell sequencing data. Our model, trained on large-scale datasets, identifies key regulators responsible for the gene program of each cell type. Additionally, our approach infers gene-specific non-linear response functions that capture complex combinatorial regulations. Moreover, we applied our model to single-cell multiomics data of mouse embryonic stem cells, enabling a deeper quantitative understanding of gene expression dynamics throughout the cell cycle. Specifically, we estimated chromatin accessibility dynamics during cell cycle progression, cell-cycle dependent transcription and degradation rates for each gene, and identified key transcription factors driving the observed transcriptional dynamics.
In conclusion, our approach provides a powerful tool for analyzing and interpreting single-cell sequencing data, enabling deeper insights into the mechanisms of gene regulation.
About the speaker: Nacho Molina is a CNRS researcher and the group leader of the Stochastic Systems Biology Lab at the IGBMC in Strasbourg. With a background in theoretical physics, he pursued a Ph.D. in computational biology at the University of Basel where he received outstanding training in Bayesian statistics, machine learning, and gene regulation. After his PhD, he underwent postdoctoral training at EPFL where he developed a novel method combining stochastic processes with hidden Markov models to analyze transcriptional bursting in individual mammalian cells. Currently at IGBMC, the main research focus of his team lies at the interface between deep learning and biophysics, combining tools from both fields to develop mechanistic and interpretable large-scale models of gene regulation. This approach allows to analyze and integrate single-cell sequencing and imaging data and generate testable predictions based on causal mechanisms. Recently, the team has started a new line of research leveraging the strength in modeling gene expression dynamics, to understand the interplay between cell cycle regulation, pluripotency maintenance, and cell differentiation.
Presentation content:Click here to download the content of this talk
Abstract: Gene expression varies between individuals and corresponds to a key step linking genotypes to phenotypes. However, our knowledge regarding the species-wide genetic control of protein abundance, including its dependency on transcript levels, is very limited. Here, we have determined quantitative proteomes of a large population of 942 diverse natural Saccharomyces cerevisiae yeast isolates. We found that mRNA and protein abundances are weakly correlated at the population gene level. While the protein co-expression network recapitulates major biological functions, differential expression patterns reveal proteomic signatures related to specific populations. Comprehensive genetic association analyses highlight that genetic variants associated with variation in protein (pQTL) and transcript (eQTL) levels poorly overlap (3%). Our results demonstrate that transcriptome and proteome are governed by distinct genetic bases, likely explained by protein turnover. It also highlights the importance of integrating these different levels of gene expression to better understand the genotype-phenotype relationship.
About the speaker: Joseph Schacherer obtained his PhD in 2005 in molecular and cellular biology from the Louis-Pasteur University in Strasbourg, France. Following the completion of his PhD, he joined the laboratory of Leonid Kruglyak at the Lewis Sigler institute of Integrative genomics at Princeton University (New Jersey, USA), where he began work on genomic approaches to study population genomics and intraspecies phenotypic variation. In 2007, he was appointed as assistant professor of genetics and genomics at the laboratory of Genetics, Genomics and Microbiology (UMR7156, University of Strasbourg - CNRS). In 2013, he became team-leader and brought together an experienced team of researchers with expertise in population genomics, genetics, bioinformatics and data analysis crucial to set up high-throughput sequencing and phenotyping experiments and analyse the data generated. The group’s long-term goal is to use population and functional genomics to have a better insight into the rules that govern the genotype-phenotype relationship within species. Moreover, he was laureate of the National Institutes of Health (NIH) R01 grant program (2012, 2017 and 2023) and was awarded an ERC Consolidator Grant in 2018. He also led the 1002 yeast genomes project (http://1002genomes.u-strasbg.fr/). He was nominated member of the Institut Universitaire de France in 2016. And since September 2017, he is professor of genetics and genomics at the University of Strasbourg.
Presentation content:Click here to download the content of this talk
Abstract: Neutron stars are fascinating astrophysical objects immersed in strong gravitational and electromagnetic fields of the order B~10^5-10^10T. These stars manifest themselves mostly as pulsars, emitting a timely very stable and regular electromagnetic signal with periods around P~ 1ms - 10s. Even though discovered 55 years ago, neutron stars still remain mysterious compact objects. Neutron star electrodynamics remains challenging for performing computer simulations because of the extraordinary large span in space and time scales involved in such stars. A typical ratio between the cyclotron frequency omegaB and the stellar rotation frequency Omega is omegaB/Omega ~ 10^16-10^19. Numerical schemes are far from being able to handle such huge ratio. However a global qualitative picture emerges slowly thanks to recent advances in numerical simulations. In this talk, I summarize the most fundamental theoretical aspects of pulsar magnetospheres and highlight the latest developments in simulations of pulsar magnetospheres, from the basic force-free approximation or from the ideal magnetohydrodynamics regime to more detailed particle-in-cell approaches including radiation reaction.
About the speaker: Jérôme Petri is Maître de Conférences at the Université de Strasbourg, Observatoire astronomique, member of the GALHECOS team. His research focuses on the theory and simulation of neutron star electrodynamics and high-energy radiation processes, linking recent multi-wavelength observations of these stars to state-of-the-art numerical modelling.
Presentation content:Click here to see the presentation
Abstract: We discuss questions and results pertaining to the presence of gaps in the spectrum of the adjacency matrix of certain families of finite graphs (in the limit where the size of the graph goes to infinity). We mostly focus on the construction of expanders (i.e. families of graphs with a uniform gap at the bottom of the spectrum), but also describe recent results or Bordenave-Collins or Sarnak-Kollar related to the presence of gaps elsewhere in the spectrum. We will also allude to similar questions for hyperbolic surfaces instead of graphs.
About the speaker: Nalini Anantharaman was a Professor at Université Paris-Sud from 2009 to 2014, at Université de Strasbourg from 2014 to 2022, and has been a professor at Collège de France since October 2022 on the "Spectral Geometry" chair. She directed the LabEx IRMIA and the ITI IRMIA++ from 2018 to early 2023.
Abstract: Experimental brain activity is known to show oscillations in specific frequency bands, which reflects neural information processing. For instance, strong oscillations at about 2Hz reflect tiredness and sleepiness, strong 40Hz oscillations indicate alertness. Changes of power in frequency bands indicate changes in information processing. For instance, it has been observed that strong activity about 10Hz and 2Hz emerge in electroencephalographic activity (EEG) when a subject loses consciousness in general anaesthesia. Numerical simulations of stochastic neural models have shown that such a change can be reproduced by changing the variance of external additive Gaussian uncorrelated noise. At a first glance, this is surprising since additive is not supposed to affect a system’s oscillatory activity or stability.
The presentation shows first how additive noise can affect a nonlinear system’s stability by applying stochastic center manifold analysis in non-delayed low-dimensional systems and delayed systems. Then, an extension to stochastic randomly connected network models shows that the observed effect also emerges. Applying random matrix theory together with mean-field theory demonstrates how additive noise tunes the stability and oscillatory activity in such systems. In sum, the mathematical studies provide an explanation why the brain’s oscillatory activity changes with changing experimental conditions.
About the speaker: Axel Hutt has studied physics at the University of Stuttgart supervised by Prof. Hermann Haken and worked on his PhD at the Max Planck Institute for Cognitive Neuroscience, for which he has received a Schloessmann Fellowship Award of the Max Planck Society in the year 2000. After positions at the Weierstrass Institute for Applied Analysis and Stochastics in Berlin, the Humboldt University Berlin and the University of Ottawa/Canada, he started working at INRIA Nancy Grand Est in 2007 and became Directeur de Recherche at INRIA in 2015. In 2010, Axel received an ERC Starting Grant. After a sabbatical stay at the German Weather Service for 4 years, from 2019 on he is working in the INRIA-team MIMESIS / iCube-team MLMS in Strasbourg on stochastic nonlinear dynamics of brain models in the context mental disorders.
Presentation content:Click here to see the presentation
Abstract: Meshes have become ubiquitous in our modern times, from video game environments and CGI characters for cartoons, to high precision simulation of physical phenomena in science and engineering. In spite of their usefulness in this wide variety of fields, representing meshes and generating them can be challenging. Each field will have its own needs and criteria, its own data structure. This talk will go over the main types of representations for meshes, 2D and 3D, and how they are used in various industries. The main focus will then be on a specific hexahedral mesh generation method for objects that are represented by their skeleton, to show some of the challenges present and a solution offered.
About the speaker: Paul Viville is a postdoctoral researcher at the University of Strasbourg in the computer science department, and the ICube Laboratory in the computer graphics and geometry team (IGG). He graduated with a Ph.D at the same university in late 2022. His research focuses on geometric modeling, volumetric mesh generation, and animation.
Presentation content:Click here to see the presentation
Abstract: Star clusters are dense and old groups of stars held together by self-gravitation, typically orbiting around (larger and less dense) galaxies. They are actually amongst the densest and oldest astrophysical structures of the Universe. Formed more than 10 billion years ago, they have witnessed and survived all the steps along the evolution of the galaxies around which they orbit. Therefore, they constitute unique probes of the history of the Universe, and in particular of our home galaxy, the Milky Way. Yet, the improvement of the observational techniques and ressources of the last decade has revealed that decoding the story told by clusters is more challenging than previously thought. Numerical simulations to the rescue! This problem is however notoriously complex, being the poster-child example of a multi-physics and multi-scale topic: up to 13 orders of magnitudes in time and spatial scales must be captured by simulations to reach a predictive nature. This multi-scale bottleneck can be, somehow seamlessly, overcome with increased computational power. But the multi-physics aspect, where complex hydrodynamics must be solved jointly with star-by-star gravitation is the real challenge. To date, no numerical technique allows to model jointly the formation and evolution of a star cluster together with its host galaxy, which hinders theoretical progress in this field. In this talk, after an introduction of the problem and the state of the art, I will present our curent efforts to add a collisional treatment in the Boltzmann equation to solve the star-by-star dynamics within the dense star clusters, while maintaining a much faster collision-less and hydrodynamical description for the rest of the galaxy. I will illustrate how our new method paves the way to a new generation of astrophysical simulations, by tackling long-lasting questions in the field.
About the speaker: Florent Renaud is a CNRS researcher in the GALHECOS team at Strasbourg Observatory since October 2023. After completing his PhD between Vienna and Strasbourg back in 2010, he moved to CEA-Saclay, Surrey (UK) and Lund (Sweden) for various post-doctoral appointments. He then became staff in Lund from 2019 to 2023. He recently obtained a fellowship from the Institute for Advanced Studies of the University of Strasbourg (USIAS) in 2023 and was concomitantly hired as a Research Director of the CNRS at Strasbourg Observatory. His research focuses on the formation and evolution of galaxies, their internal structure and their interstellar medium, with the aim to better understand the complete cosmic baryon cycle.
Presentation content:Click here to see the presentation
Abstract: In this talk we will discuss Euclidean Quantum Field Theory (EQFT) on the example of the harmonic oscillator. We will also explore a few classical probabilistic models, e.g. the free field and the Ornstein-Uhlenbeck process.
About the speaker: Xiaolin Zeng is a McF in mathematics at IRMA and maths department of the University of Strasbourg. He obtained his PhD at the University of Lyon 1 in 2015, was a postdoctoral fellow in Tel-Aviv before being appointed in Strasbourg in 2018. His research interests include probability, statistics, and mathematical physics.
Presentation content: Click here to download the content
Abstract: Formal Verification enables one to prove that a program does not contain any bug. In this talk, I will present state-of-the-art techniques for formally verifying the implementation of a nontrivial algorithm or data structure. First, I will explain how to formulate as a mathematical theorem a statement of the form: "this program behaves as intended". Second, I will explain how to leverage an "interactive proof assistant", a tool for developing machine-checked mathematical proofs, for reasoning about the behavior of the source code of the program. Finally, I will give a survey of complex programs that have been formally verified in the past decade, assessing the progress made since the pioneering work by Hoare-Floyd-Dijkstra in the late 60's.
About the speaker: Arthur Charguéraud is an Inria researcher, member of the ICube lab since 2016. He completed his PhD in 2010 at Inria Paris-Rocquencourt, then spent 18 months as a post-doc at the Max Plank Institute for Software Systems in Kaiserslautern. He was recruited at Inria Saclay in 2012, then moved to Strasbourg in 2016. His research is focused on program verification and program optimization.
Presentation content:Click here to download the content - Click here to see the presentation
Abstract: Kinetically constrained models are a class of mathematical models introduced by physicists to describe the behavior of glass. Each element of Z^2 is in state 0 or 1, and can change state at random times, but only when a constraint of the form "there are enough zeros around the element" is satisfied. There is an infinite number of possible constraints, and the properties of a model sharply depend on the constraint chosen. Therefore a very important question is that of universality: can this infinite number of models be divided into a finite number of classes depending on their behavior ? The answer is yes, and in this talk we will explain how.
About the speaker: Laure Marêché has been an assistant professor in mathematics at Université de Strasbourg and a researcher at IRMA since 2020. She obtained her PhD in 2019 at Université Paris Diderot. After that, she did a post-doc at École Polytechnique Fédérale de Lausanne. She works in the field of probability, more precisely in statistical mechanics.
Presentation content:Click here to see the presentation
Abstract: Since their introduction into cosmology, Minkowski functionals have been applied to various problems regarding the morphology of the large-scale structure of the Universe, but also the smaller-scale objects such as galaxy clusters and dark matter halos. Minkowski functionals represent a complete family of morphological measures describing the content, shape and connectivity of the underlying density field. In this talk, I will give a short comprehensible introduction into integral geometry alongside the interpretation of the Minkowski functionals in 3 dimensional Euclidean space. I will then show some applications of all four Minkowski functionals in different configurations, namely the distribution of observed galaxies in our Universe, and the distribution of matter in simulations with different initial conditions or ingredients. These morphological descriptors incorporate correlations of arbitrary order and therefore provide a complementary look at large-scale structure that could potentially reveal the presence of (primordial) non-Gaussianities, but also provide constraints on poorly known baryonic processes such as stellar or black hole feedback.
About the speaker:Katarina Kraljic is a CNRS researcher in the GALHECOS team at Strasbourg Observatory since 2022. After finishing her PhD at CEA-Saclay in 2014, she moved to Marseille for a post-doctoral appointment. She then spent 3 years as a postdoctoral fellow at the Royal Observatory in Edinburgh, before returning to Marseille in 2020 and then Strasbourg in 2022. Her research focuses on the formation and evolution of galaxies, the large-scale structure of the Universe, the internal structure of galaxies and their interstellar medium, with the aim to better understand the complete cosmic baryon cycle, both from a theoretical and observational perspective.
Presentation content: Click here to see the presentation - Click here to download the content
Abstract: L’Eau, ressource indispensable à la vie sur Terre, est de nos jours, sujette à de multiples pressions naturelles et anthropiques qui modifient voire altèrent ses états. Répondre en qualité et quantité aux besoins en eau de l’humanité sans nuire à la pérennité de la ressource et des écosystèmes nécessite une gestion intégrée à une échelle spatiale judicieusement déterminée. Face à l’urgence climatique inspirée des recommandations du GIEC (2013) et considérant l’accélération des transformations urbaines, les collectivités territoriales mènent de front des stratégies d’adaptation et d’atténuation. Ces stratégies intègrent l’implémentation des Solutions Fondées sur la Nature (SFN) qui sont des dispositifs relevant de l’ingénierie écologique. La compréhension de la trajectoire de ces SFN, des processus en jeu, de leurs fonctions écologiques et des services qu’elles offrent sont au cœur des recherches que je présenterais. Après des généralités, je ferais un focus sur différentes applications pour lesquels des modèles mathématiques ont permis de clarifier la compréhension des processus clés au sein des écosystèmes complexes.
About the speaker:Professeur en sciences de l’eau et génie de l’environnement à l’ENGEES, A. Wanko est co-responsable de l’équipe Mécanique des Fluides d’ICube et animateur de la thématique « Eau & durabilité » de la Zone Atelier Environnemental Urbaine (Zaeu), labellisée par l'Institut écologie et environnement (INEE) du Centre national de la recherche scientifique (CNRS). Thèmes de recherche : transferts de matière en milieux poreux, modélisation du couplage hydrodynamique et transferts réactifs en ingénierie écologique, phytoremediation des eaux, phytoremediation des sols, hydrologie urbaine durable.
Presentation content: Click here to see the presentation - Click here to download the content
Abstract: In this talk, I will present a (non-exhaustive) review of the state-of-the-art of our current standard cosmological model. I will show how we manipulate stochastic (quasi-)Gaussian random fields living on a curved spacetime and how we evolve them through cosmic history. I will then show that for the final stages of the cosmic evolution, when the evolution becomes too non-linear, a numerical approach prevails. I will then present you how some of the (high performance computing) simulations that we are are currently developing at the observatory, with non-Gaussian initial conditions, could (perhaps) be a game changer for the field of galaxy formation.
About the speaker:Clément Stahl is a postdoc in the GALHECOS team at Strasbourg Observatory since 2021. He obtained his PhD at the University of La Sapienza (Rome) in 2017. Before, he was a postdoctoral fellow in Valparaiso (Chile) and at the Laboratoire Astroparticules & Cosmologie (APC, Paris). His research interests include inflation, large scale structure, dark matter and dark energy both from a theoretical and numerical point of view.
Presentation content: Click here to download
Abstract: I will start with the well-known subject of addition chains, which describe the best possible ways of raising an element to a certain power, in any group (or even monoid). These are used by all computer algebra packages, when you request them to compute a high power of a large matrix, for example. Then I will say a word about the more general problem of computing in other groups, facing questions such as: when A and B are matrices, how many multiplications do you need to compute ABABABABABABABABABABABABABABABAB ? (Answer: just 4.) Very little is known in general. Finally, I will explain how similar ideas can be used to describe the "assembly index" of any object, as was recently exploited by chemists. It turns out that the assembly index of molecules can be measured experimentally rather than computed, and this opens up the possibility of detecting the presence of life on other planets. Accordingly, the talk will end with pictures of aliens.
About the speaker:Pierre Guillot is a reader in mathematics at the university of Strasbourg, and a researcher at IRMA. He obtained his PhD at the university of Cambridge (UK) in 2004, was a postdoctoral fellow in Lille and Nice before being appointed lecturer at Strasbourg. His research interests include group cohomology, Galois theory, algebraic topology and computational algebra.
Presentation content : Click here to download
Abstract: Mathematical morphology is a discipline that was developed in 1964 by Georges Matheron and Jean Serra, with the objective of characterizing and quantifying objects and structures in images by the means of sets, and more generally complete lattices. Since then, mathematical morphology can be seen as a convenient toolkit providing many operators to solve practical image processing problems. In this context, connected operators are a class of operators having the property to perserve image contours: a contour is either retained or removed, but cannot be moved.
In this talk, I will present hierarchical data structures enabling to design and compute efficiently connected operators. In particular, the max-tree, or component-tree, is a structure that is build from the level-sets of an image by storing in its nodes the connected components of the threshold sets and their inclusion relationship in its edges. On this basis, I will present an extension of this concept to color and multivalued images that we have called "component-graph". I will also present some recent results in the field of semantic segmentation involving max-trees, deep-learning and topological constraints. These concepts will be illustrated on biomedical applications.
About the speaker:Benoît Naegel is a professor in Computer Science. He obtained his PhD in the field of mathematical morphology applied to medical image analysis in 2004. From 2005 to 2007 he was a post-doctoral fellow at the University of Applied Sciences of Western Switzerland in Geneva where he worked on several biomedical projects in collaboration with the Hospital of Geneva. From 2007 to 2011 he was with the LORIA laboratory, Nancy, in the QGAR team dedicated to document processing. He is a member of ICube laboratory since 2011 where he conducts research in mathematical morphology and deep learning applied to biomedical imaging.
Abstract: The symplectic non-squeezing theorem, discovered by Gromov in 1985, has been the first result showing a fundamental difference between symplectic transformations, which form the mathematical framework for classical mechanics, and volume preserving ones. A similar but more subtle phenomenon in contact topology, the odd dimensional cousin of symplectic topology, has been found by Eliashberg, Kim and Polterovich in 2006, and refined by Fraser in 2016 and Chiu in 2017: in this case non-squeezing depends on the size of the domains, and only appears above a certain quantum scale. In my talk I will describe these fundamental results of symplectic and contact topology, briefly mentioning their relation to classical and quantum mechanics.
About the speaker: Sheila (Margherita) Sandon is a CNRS junior researcher at IRMA. She graduated from IST Lisbon with a Ph.D. degree in 2009. After a postdoctoral stay at the University of Nantes and a 2 year visit at the UMI-CNRS of the University of Montreal, she joined IRMA in 2014. She works in symplectic and contact topology.
Presentation content:Click here to see the presentation
Abstract: A large observational effort is currently underway to better probe and understand the first billion years of the Universe, an epoch from which we know very little, because of the enormous distances involved, making its observation very difficult. The new observational means involve giant radio telescopes on the ground, as well as new facilities in space (e.g., the JWST), focusing on the intergalactic medium and forming galaxies, respectively, to give us a new, complementary view of the early universe. To prepare and interpret these upcoming data, numerical simulations of the early Universe are required, involving a wide range of physics and requiring a variety of numerical approaches. I will showcase the recent works of our group in this field, including the largest simulation of the Epoch of Reionization ever made, Cosmic Dawn III (https://coda-simulation.github.io/), performed on Summit at Oak Ridge Leadership Computing Facility, the challenges it presented, and some of the challenges ahead.
About the speaker: Pierre Ocvirk is Astronome Adjoint at Observatoire de Strasbourg and member of the CDS (Centre de Données de Strasbourg). His research focuses on the modelling of the epoch of reionization using giant simulations, and galaxy formation with an emphasis on hydro-radiative processes.
Presentation content: Click here to see the presentation
Abstract: The first part will consist of a short introduction to the field of computer verified proofs. Why verify proofs with a computer? what can be verified? what are the challenges? In a second part, I will give an overview of our formalizations concerning the foundations of geometry based on the axiom systems of Hilbert, Tarski and Euclid. In particular, I will present a joint work with Michael Beeson and Freek Wiedijk which consisted in mechanically verifying the proofs of the first book of Euclid's Elements while trying to stay as close as possible to the original proofs. I will discuss the classification of 34 versions of the axioms of parallels (joint work with Pierre Boutry). We will finish by presenting a syntactic proof of the independence of the axiom of parallels (joint work with Michael Beeson and Pierre Boutry).
About the speaker: Julien Narboux is Maître de Conférences at the Université de Strasbourg, and pursues his research at ICube (IGG team). His research interests include formal proofs and formalisation of foundations of geometry and applications. He is especially interested in automated reasoning (in particular in geometry) and using proof assistants for teaching.
Presentation content: Click here to see the presentation
Abstract: We will start the talk by recalling classical facts from Morse theory. In other words we will discuss the following classical question: what do the critical points of a function on a manifold tell us about the topology of the manifold? After that we will discuss the following question. Given a manifold M and a smooth map f from M to the circle, can we deform f to a map which has no critical points? Or a minimal number of critical points ?
About the speaker: Pierre Py is a CNRS junior researcher at IRMA. He graduated from ENS Lyon with a Ph.D. degree in 2008. After a postdoctoral stay at the University of Chicago, he joined IRMA as a CNRS researcher in 2011. His research interests evolved over time from the study of symplectic diffeomorphism groups towards Kähler geometry and its interactions with topology and geometric group theory.
Abstract: I will present the state of the art on theoretical/numerical modelling of the magnetic reconnection process that is believed to be the central mechanism at work to explain magnetic eruptions in many astrophysical plasmas. In particular, I will highlight the main limitations when using standard numerical schemes to integrate the relevant set of partial differential equations in the magnetohydrodynamic framework. Finally, I will present hope for future numerical strategy based on machine/deep learning trough physics-informed neural networks.
About the speaker: Hubert Baty is Maître de Conférences at the University of Strasbourg, and pursues his research at the Observatoire astronomique de Strasbourg. His research focuses on instabilities and magnetic reconnection in magnetically-dominated plasmas with applications to solar/stellar corona and astrophysical jets.
Presentation content: Click here to see the presentation
Abstract: Numerical simulations inevitably requires the use of modern visualization methods at different stages to analyze datasets, extract information from them, to guide phenomenon modeling, to validate or invalidate models or as a tool for evaluating experimental results. The access to increasingly powerful computing machines enables scientists to simulate ever larger and more complex phenomena. Large-scale simulations generally output time-varying multivariate volumetric data, modeled by volume meshes of increasingly complex size, topology, geometry, composition, ... Direct volume rendering (DVR) is a well known method for visualizing volume data and its implementation on graphics processors (GPU), based on volume ray-casting algorithm, offers good rendering quality combined to good performance. However, such an implementation on simulated data presenting above-mentioned characteristics is a difficult problem that remains open. A key challenge of research is to make visualization techniques follow up with this drastically increasing complexity.
After an introduction to volume rendering on GPU and its adaptation to large datasets, I will address the challenges of in-situ visualization of large and complex unstructured meshes from numerical simulation through the presentation of the ANR LUM-Vis project.
About the speaker: Jonathan Sarton is an associate professor at the University of Strasbourg in the computer science department, and the ICube Laboratory in the Computer Graphics and Geometry team (IGG). His research focuses on high performance scientific visualization, volume rendering on GPU, parallel rendering, and in-situ visualization in HPC environment. He is the scientific leader of the LUM-Vis ANR project.
Abstract: Strongly-correlated quantum systems are often extremely fragile and notoriously hard to control, which poses challenges for possible technological applications. That is why a certain subclass of quantum states, the so-called topological phases of matter, recently attracted much attention. These are characterised by a certain degree of stability and robustness under perturbations, rooted in their special mathematical properties. Apriori, it is not always clear whether a given quantum state of matter is topological or not. We propose a mathematical criterion, which we call “the geometric test", to tell whether a state of matter is in a topological phase. We then apply our test to strongly-interacting states of matter in Quantum Hall effect, observed in certain 2d materials (Gallium-arsenide, graphene, ...) at low temperatures and in strong magnetic fields. I will explain the idea of the test (which works pretty well) and the results, based on recent work with Dimitri Zvonkine (CNRS, Versailles Mathematics Laboratory, Paris-Saclay University, France).
About the speaker: Semyon Klevtsov obtained his PhD in 2009 at Rutgers University (USA), working on mathematical aspects of string theory. After a post-doctoral stays in Brussels and Cologne, he joined the Institute for Advanced Mathematical Research (IRMA) at the University of Strasbourg, as professor of mathematical physics. His most recent research is focused on mathematical aspects of the strongly correlated electron systems in condensed matter physics.
Presentation content: Click here to see the presentation
Résumé
: The MLIR compiler framework is a novel compiler infrastructure that eases the process of developing new interacting compiler passes, built on top of the LLVM compiler. According to mlir.llvm.org, it especially "significantly reduces the cost of building domain specific compilers". I will shortly introduce MLIR and explain how to generate optimized compiled code using this framework.
Then, I will present our experience in the MICROCARD European project (https://microcard.eu/), in collaboration with INRIA Bordeaux and KIT among other partners. Our aim is to write software to simulate cardiac electrophysiology using whole-heart models with sub-cellular resolution, on future exascale supercomputers. It builds on the existing open source openCARP project (opencarp.org), a cardiac electrophysiology simulator for in-silico experiments. OpenCARP includes a solver and the ionic model component describing ionic transmembrane currents, as ordinary differential equations (ODEs). They are provided using a DSL (domain specific language) for ODEs named easyML. The easyML input is analyzed and transformed into code by a python parser, which we modified to plug it to MLIR and generate OpenMP and vectorized efficient code. MLIR can also be used to generate GPU code, and we plan to experiment with this in the near future.
A propos de l'orateur : Vincent Loechner is assistant professor at University of Strasbourg in the computer science department, and the ICube Laboratory in the parallelism team (ICPS). He is also part of the INRIA CAMUS team. He is in charge of the compilation and code optimization work-package of the MICROCARD European project.
Résumé : The physical processes driving galaxy formation and evolution span a vast range of scales, from the large scale structures of the universe to the turbulent interstellar medium and the interactions between light and matter. In this talk, I will present some of the empirical scaling relations that guide our understanding of these complex physical processes, focussing notably on the fate of gas within galaxies, star formation, and the relation between galaxies and their surrounding dark matter haloes.
A propos de l'orateur : Jonathan Freundlich did his Ph.D. between 2012 and 2015 at the Paris Observatory, probing star formation across cosmic time and modelling the influence of baryons on dark matter haloes. He was afterwards a postdoc at the Hebrew University of Jerusalem, where he gained experience in analysing cosmological simulations. He became Maître de Conférences at the Strasbourg Observatory in 2021, within the Galaxies, High Energy, Cosmology, Compact Objects & Stars (GALHECOS) research group.
Résumé : In this talk, we want to introduce different examples or problems of interaction between deep learning, geometry, PDEs and numerical methods. We will start by illustrating the ability of deep learning to deal with physical problems starting from a classical problem in fluid mechanics: the closure. In a second step, we will introduce recent works mixing learning and differential geometry which allow to tackle unstructured data. We will illustrate this with simple examples from PDEs. Finally, we will show how ideas from analytical mechanics (and symplectic geometry) can interact with machine learning and numerical simulations of PDEs.
A propos de l'orateur : Emmanuel Franck did his Ph.D. thesis from 2009 to 2012 at the CEA on the numerical approximation of the radiative transfer equation. After that he did a 2 year post-doc at the Max Planck Institute for Plasma Physics in Munich on numerical methods for MHD in nuclear fusion. He is an INRIA young researcher since 2014 and his work focuses on the numerical approximation of PDEs in fluid mechanics and plasma physics. He is a member of the Modelling and Control (MOCO) research group at IRMA, Strasbourg.
