HAL-Ecole des Ponts ParisTech
Not a member yet
42743 research outputs found
Sort by
Learning Representations on the Unit Sphere: Investigating Angular Gaussian and von Mises-Fisher Distributions for Online Continual Learning
No full textInternational audienceWe use the maximum a posteriori estimation principle for learning representations distributed on the unit sphere. We propose to use the angular Gaussian distribution, which corresponds to a Gaussian projected on the unit-sphere and derive the associated loss function. We also consider the von Mises-Fisher distribution, which is the conditional of a Gaussian in the unit-sphere. The learned representations are pushed toward fixed directions, which are the prior means of the Gaussians; allowing for a learning strategy that is resilient to data drift. This makes it suitable for online continual learning, which is the problem of training neural networks on a continuous data stream, where multiple classification tasks are presented sequentially so that data from past tasks are no longer accessible, and data from the current task can be seen only once. To address this challenging scenario, we propose a memory-based representation learning technique equipped with our new loss functions. Our approach does not require negative data or knowledge of task boundaries and performs well with smaller batch sizes while being computationally efficient. We demonstrate with extensive experiments that the proposed method outperforms the current state-of-the-art methods on both standard evaluation scenarios and realistic scenarios with blurry task boundaries. For reproducibility, we use the same training pipeline for every compared method and share the code at https://t.ly/SQTj
Trends and Inequality in Lifetime Earnings in France
No full textThis paper is the first to compute lifetime earnings (LTE) in France for a large number of cohorts entering the labour market between 1967 and 1987, and to analyze their main determinants, as well as those of the evolution of the gender gap in LTE. We compare our results with evidence by Guvenen et al. (2022a) for the US, documenting sharp differences between the two countries. Median LTE show similar flat trends in both countries, but in France this results from a moderate increase for both genders together with increased female participation, while in the US, LTE declines for men and sharply grows for women. There have been marked changes in age profiles, as for both genders younger cohorts have experienced a decrease in entry wages that has been more than offset by faster wage growth. Our analysis of inequality finds that it is lower when we focus on LTE than in the cross-section, and that it follows a U-shaped pattern, although the increase is much smaller in France than that observed in the US. Lastly, we also find that i) education (returns and changes in attainment) plays a key role in shaping LTE across cohorts, ii) place of birth has a large influence on lifetime earnings and iii) differences in working time explain an increasing part of the gender gap in LTE over time as both men and women have increased the number of years they work but women have done so largely through part-time employment
Going beyond carbon: Influence of structural parameters on the environmental impacts of typical building structures
No full textInternational audienceThe construction sector has a large impact on the environment, be it climate change, biodiversity loss or resources depletion. Part of those impacts comes from construction materials, and more specifically, concrete and steel building structures. Early-stage design tools including environmental damage assessment are therefore required to achieve sustainable construction practices. This article presents a parametric approach aiming at studying the influence of early-stage structural parameters on the environmental impacts of a structure. The developed methodology combines parametric structural design with a multicriteria cradle-to-grave Life Cycle Assessment, applied to typical housing structures such as grid based beam column structures with a central core for lateral stability. Results show that span, number of levels and materials greatly influence the environmental impacts of considered structures. Climate change scores per floor area range from 80 kgCO2/m 2 for short span timber structure to 215 kgCO2/m 2 for long span steel buildings with few levels. Beams and slabs have the largest contribution in such impacts. Changing the production processes of materials is a way to reduce greenhouse gas emissions. However, decarbonizing steel production processes or reducing the cement content of concrete participate in burdenshifting with increased scores in several impact categories, such as carcinogenic toxicity or ozone depletion
Utilisation des jumeaux numérique des bâtiments, l'IA et l'IoT pour l'optimisation des services orientés localisation indoor
No full textIntegrating Building Information Modeling (BIM) with the Internet of Things (IoT) aims to generate the digital twin of a building, providing a real-time representation of its physical counterpart. Indeed, BIM incorporates a 3D model of a building and a database associated with each object of this model. These static data, such as the material of the objects, can be useful for optimising the deployment of an IoT network in terms of coverage, and can be merged with the data collected in real-time by this IoT network to generate the building’s digital twin. This concept has become a powerful tool, particularly in Industry 4.0, for enhancing the generation and management of efficient location-based services through a real-time interactive model reflecting the building's current state. This thesis presents a "BIM-based" method for generating the digital twin of a building and implementing a robust location-based service. This process includes the optimal deployment of an IoT network in terms of coverage adapted to the service, and the construction of an IoT network architecture using users' devices to guarantee robust continuity of service while extending the service to devices without Internet access. First, the optimal deployment algorithm is proposed with different inputs: a theoretical map, a standardised evacuation plan, and the 3D BIM. This process is based on a building skeletonisation technique and a multi-objective genetic algorithm. This algorithm enables optimal 2D and 3D deployment using the BIM, for radio and visual coverage, according to connectivity, Full-View and multilateration constraints. These constraints relate to one of the indoor location-based services use case, which is indoor guidance. Implementing these innovative services requires a network of IoT objects deployed in the building, which can lead to high initial costs. It is shown by simulation that our proposal outperforms state of the art techniques in all scenarios and according to the metrics considered. Once the network has been deployed, a multi-tier architecture for IoT networks is proposed. This enables the service to be maintained in an unstable topology, where devices can appear and disappear unpredictably. "Dominant" devices are elected as relays for the remote server using a selection process. Two variants of this process are studied: the first is based on graph theory, while the second is based on reinforcement learning. The conducted performance evaluation shows that this architecture enables continuity of service in an aggressive topology, whereas a centralized architecture results in network congestion in the same scenario. Therefore, location-based service facing a high density of devices, such as indoor guidance, which can lead to overhead and interference can employ this architecture to improve the quality of service (QoS).The thesis also evaluates the overall process in terms of its carbon footprint. A model of this metric has been proposed, firstly by studying the contribution of BIM to deployment optimization, and then by comparing the carbon impact of data transport for decentralized and centralized networks. The efficiency of the process is demonstrated by numerical analysis, with a reduction of an order of magnitude ranging from 10^2 to 10^5 kgCO2eq/year. As a practical outcome, the thesis presents BOMNET, an Open Source plug-in allowing BIM integration in the OMNeT++ software for realistic signal emulation. BOMNET's contribution has been demonstrated by comparing the results with a simulation without BIM and to reality. BOMNET reduces the absolute error between simulation and reality by up to 70%L'intégration du BIM (Building Information Modeling) avec l'Internet des Objets (IoT) vise à générer le jumeau numérique d'un bâtiment, offrant une représentation en temps réel de son homologue physique. En effet, le BIM comporte un modèle 3D d'un bâtiment et une base de données associée à chaque objet du modèle. Ces données statiques peuvent permettre d'optimiser le déploiement d'un réseau IoT en termes de couverture, et être fusionnées avec les données collectées en temps réel par ce réseau pour générer le jumeau numérique du bâtiment. Ce concept s'est révélé être un outil important, particulièrement dans l'industrie 4.0, améliorant la génération et la gestion de services basé sur la localisation indoor efficaces et robustes. Dans cette thèse, une méthode de génération du jumeau numérique d'un bâtiment et la mise en place d'un service de proximité robuste à partir du BIM sont présentées. Ce procédé inclut le déploiement optimal d'un réseau IoT en termes de couverture adaptée au service, puis la construction d'une architecture de réseau IoT utilisant les appareils des usagers pour garantir une continuité de service robuste et étendre l'accès au service aux appareils sans accès internet.Tout d'abord, l'algorithme de déploiement optimal est proposé avec différentes entrées : une carte théorique, un plan d'évacuation normalisé, et le BIM 3D. Cet algorithme s'appuie sur une méthode de squelettisation du bâtiment et un algorithme génétique multi-objectif. Ce procédé permet le déploiement 3D optimal à partir du BIM pour la couverture radio et visuelle, selon les contraintes de connectivité, de Full-View et de multilatération. Ces contraintes concernent l'un des cas d'usage des services orientés localisation indoor, à savoir le guidage en intérieur. La mise en œuvre de ces services innovants nécessite un réseau IoT déployé dans le bâtiment, ce qui peut entraîner des coûts initiaux élevés. Il est démontré par simulation que la contribution est plus performante que l'état de l'art.Une fois le réseau déployé, une architecture multi-tier pour réseau IoT est proposée. Celle-ci permet de maintenir le service dans une topologie instable, où les appareils peuvent apparaître et disparaître de manière imprévisible. Des appareils "dominants" sont élus comme relais du serveur via un processus de sélection. Deux variantes de ce processus sont étudiées, respectivement basées sur la théorie des graphes et sur l'apprentissage par renforcement. L'évaluation des performances réalisée démontre que cette architecture permet la continuité de service dans une topologie agressive, alors qu'une architecture centralisée résulte en la congestion du réseau dans le même scénario. Par conséquent, les services basés sur la localisation indoor confrontés à une forte densité d'appareils, tels que le guidage indoor peuvent utiliser cette architecture pour améliorer la qualité de service.La thèse évalue également le processus global en termes d'empreinte carbone. Une modélisation de cette métrique a été proposée, tout d'abord en étudiant l'apport du BIM dans l'optimisation de déploiement, puis en comparant l'impact carbone due au transport des données des réseaux décentralisés et centralisés. L'efficacité du processus est démontrée par analyse numérique, avec une réduction allant de 10^2 à 10^5 kgCO2eq/an. Comme résultat pratique, la thèse présente BOMNET, un plug-in Open Source permettant l'intégration de BIM dans le logiciel OMNeT++ pour une émulation réaliste des signaux. L’apport de BOMNET a été démontré en comparant les résultats d'une simulation sans BIM et de mesures réelles. BOMNET permet de réduire l’erreur absolue entre la simulation et la réalité jusqu’à 70
Achats transfrontaliers de carburant à la frontièrefranco-allemande
No full textThis paper exploits the introduction of the German carbon tax in 2021 as well as excise tax rebates on fuel in France and in Germany, consecutive to the 2022 oil crisis, to infer how fuel tax revenue responds to changes in relative prices. Based on French high-frequency transaction-level data issued from individual banking accounts, we find substantial displacement between foreign and domestic consumption. When relative prices increase by 1%, the relative cross-border demand decreases by 7.8%. Moreover, there is no substantial difference in demand response to either carbon or excise taxes.Cet article exploite l’introduction de la taxe carbone allemande en 2021 ainsi que les réductions de taxes d’accise sur les carburants en France et en Allemagne, consécutives à la crise pétrolière de 2022, pour déduire comment les recettes des taxes sur les carburants réagissent aux changements des prix relatifs. Sur la base des données françaises à haute fréquence issues des comptes bancaires individuels, nous constatons un déplacement substantiel entre la consommation étrangère et la consommation nationale. Lorsque les prix relatifs augmentent de 1 %, la demande transfrontalière relative diminue de 7,8 %. En outre, il n’y a pas de différence substantielle dans la réponse de la demande aux taxes sur le carbone ou aux taxes d'accise
Effect of thermal cycles on sand-concrete interface under constant shear stress
No full textInternational audienceEnergy geostructures provide a profitable solution for structural support as well as the heating and cooling of buildings. However, their activation can produce in structure and its surrounding soil thermally-induced stresses and strains with a crucial role played by the soil-structure interface which is the thin layer that transmits the thermal and mechanical loads. In the present work, the thermo-mechanical behaviour of the soil–structure interface is investigated using a temperature-controlled direct shear box. Tests were conducted on loose and dense sands in contact with a concrete plate representing the structure's surface. After applying the normal stress (25, 50, or 100 kPa), a shear stress was increased up to 45% of the shear strength. This stress state was then maintained while 20 cycles of temperature varying between 13 °C and 34 °C at the soil-concrete interface were applied. The obtained experimental results reveal that temperature cycles lead to irreversible cumulative displacements in both normal and shear directions. These displacements were found to be more significant in loose sand compared to dense sand. Additionally, under higher stress levels, greater displacements were observed for both soil densities
The trade-off between socio-environmental awareness and renewable penetration targets in energy transition roadmaps
No full textInternational audienceRecent years have seen Sustainable Development Goals mark the way towards a sustainable future and put the growth of renewable energy sources in the spotlight. This increase, led by PV and wind technologies, will be necessary to meet renewable penetration targets in upcoming decades. However, these sources require vast amounts of land, a limited resource affecting other Sustainable Development Goals such as ecosystem protection or food production. This article explores the land–energy nexus through the minimization of the system total cost that incorporates a form of suitability for land occupation. Three differentiated regimes appear under a changing suitability cost, with a gap of 70% penetration between the two extremes, and with veryhigh suitability costs unable to meet future objectives given their cap at around the 35% penetration mark. The intermediate approach presents the best fulfillment of Sustainable Development Goals overall, with an increase between 15 and 30% with respect to the other regimes. Existing development plans for the Spanish system point towards this midway approach, a sign of their commitment to sustainability. This study shows that a balanced consideration of land and energy should be prioritized, as the negative effects that arise in a biased approach outweigh the benefits
Estimation of statistics of transitions and Hill relation for Langevin dynamics
No full textInternational audienceIn molecular dynamics, statistics of transitions, such as the mean transition time, are macroscopic observables which provide important dynamical information on the underlying microscopic stochastic process. A direct estimation using simulations of microscopic trajectories over long time scales is typically computationally intractable in metastable situations. To overcome this issue, several numerical methods rely on a potential-theoretic identity, sometimes attributed to Hill in the computational statistical physics litterature, which expresses statistics of transitions in terms of the invariant measure of the sequence of configurations by which the underlying process enters metastable sets. The use of this identity then allows to replace the long time simulation problem with a rare event sampling problem, for which efficient algorithms are available. In this article, we rigorously analyse such a method for molecular systems modelled by the Langevin dynamics. Our main contributions are twofold. First, we prove the Hill relation in the fairly general context of positive Harris recurrent chains, and show that this formula applies to the Langevin dynamics. Second, we provide an explicit expression of the invariant measure involved in the Hill relation, and describe an elementary exact simulation procedure. Overall, this yields a simple and complete numerical method to estimate statistics of transitions
Second RILEM International Conference on Earthen Construction
No full textInternational audienceThis book gathers the peer-reviewed papers presented at the Second RILEM International Conference on Earthen Construction (ICEC), held in Edinburgh, United Kingdom, on July 8–10, 2024. It highlights the latest advances and innovations in the field of con earth-based building materials and construction. The conference topics encompass material characterisation and quality control, hydro-mechanical behaviour, reinforcement behaviour, seismic behaviour, in situ and field testing, additive manufacturing (3D printing), rheology, biostabilisation, molecular simulation, microstructure, durability, fire performance, hygro-thermal behaviour, life cycle analysis, climate change adaptation, economic impacts, and earthen architecture. As such, the book represents an invaluable, up-to-the-minute tool, and offers an important platform to engineers, architects, and geophysicists
Contribution du LMD à l’étude du rôle des nuages dans le changement climatique
No full textInternational audienceIn the early 1990s, the role of clouds in climate change became a central issue in the climate sciences. This article discusses the LMD's role in advancing this theme at international level: in highlighting the role of low clouds in uncertainties about climate sensitivity, in understanding the feedback mechanisms associated with low clouds and anvil clouds, and then in reducing these uncertainties using space observations and field campaigns. The LMD's role in this area is characteristic of its culture, which is to understand climate by studying processes, and vice versa.Au début des années 1990, le rôle des nuages dans le changement climatique devient une question centrale dans les sciences du climat. Cet article discute du rôle du LMD dans l’avancée de cette thématique au niveau international : dans la mise en évidence du rôle des nuages bas dans les incertitudes sur la sensibilité climatique, dans la compréhension des mécanismes de rétroaction associés aux nuages bas et aux nuages d’enclume, puis dans la réduction de ces incertitudes à l’aide d’observations spatiales et de campagnes de terrain. Le rôle du LMD dans cette thématique est caractéristique de sa culture, à savoir comprendre le climat par l’étude des processus, et vice-versa