HAL-Ecole des Ponts ParisTech
Not a member yet
42743 research outputs found
Sort by
Les grands réseaux dans et pour la transition socio‑écologique : un nouvel âge des infrastructures ?
No full textInternational audienceCet article rappelle d’abord que les grands réseaux techniques d’énergie, d’eau, de transport et de communication sont, depuis deux siècles, une composante essentielle de la modernité occidentale dans ses principales dimensions. Au moment où celle ci fait l’objet de mises en question multiples, notamment en lien avec les changements environnementaux planétaires constitutifs de l’anthropocène, la question du futur des grands réseaux est posée. L’article explore les fondements et les conditions possibles de futurs infrastructurels alternatifs plus compatibles avec le respect des frontières planétaires
Meeting climate target with realistic demand-side policies in the residential sector
No full textInternational audienceAbstract The EU has established an ambitious policy framework for demand-side mitigation in buildings towards net-zero targets. Here, we conduct a comprehensive quantitative assessment of 384 demand-side policy combinations for residential space heating that complement supply-side decarbonization efforts. We show that the implementation of EU Emissions Trading System 2, even when combined with deep decarbonization of energy supply, falls short of climate targets. Beyond ETS 2, we emphasize the need for ambitious subsidies for heat pumps as a critical component of a successful strategy. Conversely, a large-scale generic ‘Renovation Wave’ modestly contributes to decarbonization, is not a cost-effective strategy at the EU level and requires significant increases in public spending. We advocate for the implementation of a carbon tax, paired with substantial subsidies for heat pumps and targeted incentives for home insulation by country and building. This approach supports the decarbonization of the residential sector, limits the strain on the electricity grid, and alleviates energy poverty
Impact of gas dry deposition parameterization on secondary particle formation in an urban canyon
No full textInternational audienceThis study investigates the impact of the parameterization of dry deposition on the local gas-particle partitioning between ammonium nitrate NH4NO3 and its precursor gases (ammonia, NH3 and nitric acid, HNO3) through chemistry-coupled large-eddy simulations on the dispersion of reactive gaseous and particulate pollutant in an idealized street canyon. A key factor in the parameterization of dry deposition is the effective Henry's Law constant H*. Three distinct characterizations of H* are compared: two drawn from existing literature (referred to as Model 1 and Model 2) and one typical of low pH conditions (referred to as Model 3), which could be more representative of urban areas. Model 3 shows contrasting gas-particle partitioning results between NH3, HNO3, and NH4NO3 with Model 1 and Model 2. In detail, the NH4NO3 concentrations in the street of Model 1 and Model 2 are smaller than the background NH4NO3 concentration. However, Model 3 shows higher NH4NO3 concentration in the street than the background NH4NO3 concentration. This is because the dry deposition fluxes of NH3 and HNO3 are higher in Model 1 and Model 2 than in Model 3, resulting in less available NH3 and HNO3 for NH4NO3 formation. These findings highlight the importance of selecting an appropriate H* characterization that is tailored to the specific environmental conditions under investigation
A PTAS for ℓ 0 -Low Rank Approximation: Solving Dense CSPs over Reals
No full textInternational audienceWe consider the Low Rank Approximation problem, where the input consists of a matrix A and an integer k, and the goal is to find a matrix B of rank at most k that minimizes , which is the number of entries where A and B differ. For any constant k and ε>0, we present a polynomial time (1+ε)-approximation time for this problem, which significantly improves the previous best poly(k)-approximation.Our algorithm is obtained by viewing the problem as a Constraint Satisfaction Problem (CSP) where each row and column becomes a variable that can have a value from . In this view, we have a constraint between each row and column, which results in a dense CSP, a well-studied topic in approximation algorithms. While most of previous algorithms focus on finite-size (or constant-size) domains and involve an exhaustive enumeration over the entire domain, we present a new framework that bypasses such an enumeration in . We also use tools from the rich literature of Low Rank Approximation in different objectives (e.g., with p∈(0,∞)) or domains (e.g., finite fields/generalized Boolean). We believe that our techniques might be useful to study other real-valued CSPs and matrix optimization problems.On the hardness side, when k is part of the input, we prove that Low Rank Approximation is NP-hard to approximate within a factor of Ω(logn). This is the first superconstant NP-hardness of approximation for any p∈[0,∞] that does not rely on stronger conjectures (e.g., the Small Set Expansion Hypothesis)
Topological analysis of extreme events
No full textInternational audienceNumerical and theoretical studies have shown that transient atmospheric motions leading to weather extremes can be classified through the stability of a state of a dynamical system and the instantaneous dimension [1]. While the asymptotic values of these quantities can be computed theoretically only for specific systems, their numerical counterpart for climate observables provides information on the rarity, predictability, and persistence of specific states. There is therefore both theoretical and practical interest in bridging such numerical metrics with their theoretical counterpart. In this work, we present a first attempt to relate the instantaneous dimension and other local metrics with the topological properties of the templex in the deterministic [2]. The templex provides the key characteristics of the topological structure underlying a dynamical system. This work will present results for the classical, deterministic and random Lorenz attractor [3;4]. References[1] Faranda D., Messori G., & Yiou P., Scientific reports 7(1) 41278 (2017). [2] Charó G. D, Letellier C. & Sciamarella D., Chaos: An Interdisciplinary Journal of Nonlinear Science, 32.8 (2022).[3] Lorenz, E. N.Journal of atmospheric sciences, 20(2), 130-141, (1963). [4] Ghil M. & Sciamarella D., Nonlinear Processes in Geophysics, 30(4), 399-434. (2023
A Mouse Model of Mild Clostridioides difficile Infection for the Characterization of Natural Immune Responses
No full textInternational audienceBackground: We describe a model of primary mild-Clostridioides difficile infection (CDI) in a naïve host, including gut microbiota analysis, weight loss, mortality, length of colonization. This model was used in order to describe the kinetics of humoral (IgG, IgM) and mucosal (IgA) immune responses against toxins (TcdA/TcdB) and surface proteins (SlpA/FliC). (2) Methods: A total of 105 CFU vegetative forms of C. difficile 630Δerm were used for challenge by oral administration after dysbiosis, induced by a cocktail of antibiotics. Gut microbiota dysbiosis was confirmed and described by 16S rDNA sequencing. We sacrificed C57Bl/6 mice after different stages of infection (day 6, 2, 7, 14, 21, 28, and 56) to evaluate IgM, IgG against TcdA, TcdB, SlpA, FliC in blood samples, and IgA in the cecal contents collected. (3) Results: In our model, we observed a reproducible gut microbiota dysbiosis, allowing for C. difficile digestive colonization. CDI was objectivized by a mean weight loss of 13.1% and associated with a low mortality rate of 15.7% of mice. We observed an increase in IgM anti-toxins as early as D7 after challenge. IgG increased since D21, and IgA anti-toxins were secreted in cecal contents. Unexpectedly, neither anti-SlpA nor anti-FliC IgG or IgA were observed in our model. (4) Conclusions: In our model, we induced a gut microbiota dysbiosis, allowing a mild CDI to spontaneously resolve, with a digestive clearance observed since D14. After this primary CDI, we can study the development of specific immune responses in blood and cecal contents
Estimation 3D de la posture humaine en environnement de travail à l'aide de réseaux de neurones profonds
No full textThe objectives of the thesis is to develop the methods and frameworks to analysis 3D human postures in the working environment for ergonomic propose.Ergonomics is a discipline which consists of understanding body work with the objective of preserving the health of operators while allowing achievement of the expected quality. The postures of operators at work stations are one of the factors in the appearance of occupational diseases, and the characterization of a posture is a step in the pre-diagnosis of a work situation. Artificial intelligence methods using 3D human pose estimation to detect unsuitable postures at work could help ergonomist to establish their diagnosis on a large quantity of data.This thesis proposes three works on 3D human pose estimation to attack the difficulties of unconstrained environment such as work stations.The first work proposes a synthetic 3D human pose generation algorithm for training 2D to 3D human pose lifting. We tackle with the domain gap problem between public research data which have constrained environment plus limited number of action and unconstrained working environment data with much more variety of actions. This work presents an algorithm which allows to generate synthetic 3D human skeletons on the fly during the training, following a Markov-tree type distribution which evolve through out time to create unseen poses. This work also proposes a scaleless multi-view training process based on purely synthetic data generated from a few initial poses. We evaluate our approach on two benchmark datasets and achieve promising results in a zero shot setup.The second work proposes a framework of making 3D wholebody annotations from multi-view image data as well as a few datasets and a benchmark built based on this framework to tackle with the skeleton model capability problem. Public research data normally only has around 20 keypoints which is not enough for ergonomist to measure certain aspects like supination-pronation angles, while the wholebody skeleton has 133 keypoints, capable for obtaining the necessary information. The annotation-making framework contains 3 steps from multi-view 3D geometry reconstruction, to incomplete skeleton completion, and finally hand/face refinement through diffusion. With this framework, we introduce 3 datasets as extensions of existing Human3.6M, CMU-Panoptic and MPI-INF-3DHP datasets with wholebody 2D and 3D keypoint annotations for body, face, and hands. A benchmark of three tasks is proposed based on Human3.6M wholebody extension: 3D whole-body lifting from complete 2D keypoints, from incomplete 2D keypoints, and from monocular images.The third work proposes an algorithm which allows a continuous estimation of human poses through time with very limited input frames to tackle potential corrupted video sequences in unconstrained environment where the workers are not always observed or even in the screen due to their movements. This work proposes a novel approach that models human motion as a continuous function implemented by a neural network, akin to neural implicit representations. We conduct a comprehensive comparison of this approach with state-of-the-art motion prediction methods on three popular datasets, demonstrating significant improvements over the baselines in most cases.Finally, we made a demonstration that perform 2D and 3D human pose estimation, as well as critical pose detection for ergonomic analysis, capable of fast processing even on a CPU-only computer.Les objectifs de la thèse sont de développer des méthodes et des cadres d'analyse de la posture humaine 3D en environnement de travail pour l'ergonomie. L’ergonomie est une discipline qui consiste à comprendre le fonctionnement du corps lorsqu'il travaille pour l’objectif de préserver la santé des opérateurs tout en permettant l’atteinte de la qualité attendue. Les postures des opérateurs au poste de travail sont un des facteurs d'apparition des maladies professionnelles, et la caractérisation d’une posture est une étape du pré-diagnostic d’une situation de travail. Les méthodes d’intelligence artificielle autour de l’estimation 3D de la pose humaine pour détecter les postures inadaptées au travail peuvent ainsi aider l’ergonome à établir son diagnostic sur un grand nombre de données. Cette thèse propose trois travaux autour de l'estimation de pose humaine 3D pour s'attaquer aux difficultés de mise en œuvre en environnement non contraint tels que les postes de travail.La première contribution propose un algorithme synthétique de génération de poses humaines en 3D. Nous abordons le problème de l'écart de domaine selon lequel les scénarios de travail a plus de variété d'actions et d'environnement que les données de recherche publique. Ce travail présente un algorithme qui permet de générer des squelettes humains 3D synthétiques pendant l'entraînement de réseau des neurons, suivant une distribution de type arbre de Markov qui évolue au fil du temps pour créer des nouvelles postures. Ce travail propose également un processus d'entraînement multi-vues sans échelle basé sur des données purement synthétiques générées à partir de quelques postures initiales. Nous évaluons notre approche sur les deux ensembles de données de référence et obtenons des résultats prometteurs dans une configuration sans aucune donnée réelle. Le deuxième travail propose un cadre de création d'annotations 3D du corps entier à partir d'images multi-vues ainsi qu'un benchmark construit sur la base de ce cadre. Les données couramment utilisées ne comportent normalement qu'une vingtaine d'articulations, ce qui n'est pas suffisant pour qu'un ergonome puisse mesurer certains aspects comme les angles de supination-pronation, c'est pourquoi nous proposons un squelette du corps entier compte 133 articulations, capables de contenir les informations nécessaires. Le cadre de création d'annotations contient 3 étapes allant de la reconstruction géométrique 3D multi-vues à la complétion des squelettes incomplets, et enfin au raffinement main/visage par diffusion. Avec ce cadre, nous introduisons 3 ensembles de données en tant qu'extensions des ensembles de données Human3.6M, CMU-Panoptic et MPI-INF-3DHP existants avec des annotations de points clés 2D et 3D du corps entier pour le corps, le visage et les mains. Un benchmark de trois tâches est proposé sur la base de l'extension du corps entier de Human3.6M.Le troisième travail propose un algorithme qui permet une prédiction continue des poses humaines à travers le temps avec des images d'entrée très limitées pour aborder des séquences vidéo potentiellement corrompues dans un environnement sans contrainte où les travailleurs ne sont pas toujours observés ou même à l'écran en raison de leurs mouvements. Ce travail propose une nouvelle approche qui modélise le mouvement humain comme une fonction continue mise en œuvre par un réseau neuronal, semblable à des représentations neuronales implicites. Nous effectuons une comparaison complète de cette approche avec des méthodes de prédiction de mouvement de pointe sur trois ensembles de données populaires, démontrant des améliorations significatives par rapport aux lignes de base dans la plupart des cas. Enfin, nous avons réalisé un démonstrateur qui effectue une estimation de la pose humaine en 2D et 3D, ainsi qu'une détection des poses critiques pour une analyse ergonomique, capable d'analyse rapide même sur un ordinateur équipé uniquement d'un CPU
Tax Design, Information, and Elasticities: Evidence From the French Wealth Tax
No full textUsing exhaustive administrative wealth and income tax data, we study a French wealth tax reform that scaled back information reporting requirements below a certain wealth threshold. We develop a dynamic bunching approach that permits estimating the average response to the reform, the share of compliers, and the LATE. Reported wealth declines sharply in response to the reform and annual wealth growth rates are on average 20% lower among affected taxpayers. This decline appears due to increased evasion facilitated by the lower reporting requirements, as suggested by the fall in self-reported wealth but the lack of response in third-party-reported labor and capital incomes. By contrast, the elasticities to tax rates estimated are very small and insignificant. This illustrates the critical role of information reporting policies in shaping taxpayers' behavior