25353 research outputs found
Sort by
Residue of special functions of Anderson A-modules at the characteristic graph
No full textInternational audienc
Conditional Gradient-based Textual Inversion
No full textInternational audienceGenerative models excel in image generation but often require trail-and-errors for specific concepts. Textual inversion offers a solution; yet, is computationally costly. We propose using conditional gradient data to select or sample informative timesteps for textual inversion. Our methods improve computational cost and generation quality
Optimisation multi-objectif pour l'analyse de données dans le Cloud
No full textBig data query processing has become increasingly important, prompting the development and cloud deployment of numerous systems. However, automatically tuning the numerous parameters in these big data systems introduces growing complexity in meeting users' performance goals and budgetary constraints. Determining optimal configurations is challenging due to the need to address: 1) multiple competing performance goals and budgetary constraints, such as low latency and low cost, 2) a high-dimensional parameter space with complex parameter control, and 3) the requirement for high computational efficiency in cloud use, typically within 1-2 seconds.To address the above challenges, this thesis proposes efficient multi-objective optimization (MOO) algorithms for a cloud optimizer to meet various user objectives. It computes Pareto optimal configurations for big data queries within a high-dimensional parameter space while adhering to stringent solving time requirements. More specifically, this thesis introduces the following contributions.The first contribution of this thesis is a benchmarking analysis of existing MOO methods and solvers, identifying their limitations, particularly in terms of efficiency and the quality of Pareto solutions, when applied to cloud optimization.The second contribution introduces MOO algorithms designed to compute Pareto optimal solutions for query stages, which are units defined by shuffle boundaries. In production-scale big data processing, each stage operates within a high-dimensional parameter space, with thousands of parallel instances. Each instance requires resource parameters determined upon assignment to one of thousands of machines, as exemplified by systems like MaxCompute. To achieve Pareto optimality for each query stage, we propose a novel hierarchical MOO approach. This method decomposes the stage-level MOO problem into multiple parallel instance-level MOO problems and efficiently derives stage-level MOO solutions from instance-level MOO solutions. Evaluation results using production workloads demonstrate that our hierarchical MOO approach outperforms existing MOO methods by 4% to 77% in terms of performance and up to 48% in cost reduction while operating within 0.02 to 0.23 seconds compared to current optimizers and schedulers.Our third contribution aims to achieve Pareto optimality for the entire query with finer-granularity control of parameters. In big data systems like Spark, some parameters can be tuned independently for each query stage, while others are shared across all stages, introducing a high-dimensional parameter space and complex constraints. To address this challenge, we propose a new approach called Hierarchical MOO with Constraints (HMOOC). This method decomposes the optimization problem of a large parameter space into smaller subproblems, each constrained to use the same shared parameters. Given that these subproblems are not independent, we develop techniques to generate a sufficiently large set of candidate solutions and efficiently aggregate them to form global Pareto optimal solutions. Evaluation results using TPC-H and TPC-DS benchmarks demonstrate that HMOOC outperforms existing MOO methods, achieving a 4.7% to 54.1% improvement in hypervolume and an 81% to 98.3% reduction in solving time.Le traitement des requêtes Big Data est devenu de plus en plus important, ce qui a conduit au développement et au déploiement dans le cloud de nombreux systèmes. Cependant, le réglage automatique des nombreux paramètres de ces systèmes Big Data introduit une complexité croissante pour répondre aux objectifs de performance et aux contraintes budgétaires des utilisateurs. La détermination des configurations optimales est un défi en raison de la nécessité de prendre en compte : 1) plusieurs objectifs de performances et contraintes budgétaires concurrents, tels qu'une faible latence et un faible coût, 2) un espace de paramètres de grande dimension avec un contrôle de paramètres complexe, et 3) l'exigence d'une configuration élevée. efficacité de calcul dans l'utilisation du cloud, généralement en 1 à 2 secondes.Pour relever les défis ci-dessus, cette thèse propose des algorithmes d'optimisation multi-objectifs (MOO) efficaces pour un optimiseur de cloud afin de répondre à divers objectifs des utilisateurs. Il calcule les configurations Pareto optimales pour les requêtes Big Data dans un espace de paramètres de grande dimension tout en respectant des exigences strictes en matière de temps de résolution. Plus précisément, cette thèse présente les contributions suivantes.La première contribution de cette thèse est une analyse comparative des méthodes et solveurs MOO existants, identifiant leurs limites, notamment en termes d'efficacité et de qualité des solutions Pareto, lorsqu'elles sont appliquées à l'optimisation du cloud.La deuxième contribution présente les algorithmes MOO conçus pour calculer les solutions optimales de Pareto pour les étapes de requête, qui sont des unités définies par des limites de mélange. Dans le traitement du Big Data à l’échelle de la production, chaque étape opère dans un espace de paramètres de grande dimension, avec des milliers d’instances parallèles. Chaque instance nécessite des paramètres de ressources déterminés lors de l'affectation à l'une des milliers de machines, comme en témoignent des systèmes comme MaxCompute. Pour atteindre l’optimalité Pareto pour chaque étape de requête, nous proposons une nouvelle approche hiérarchique MOO. Cette méthode décompose le problème MOO au niveau de l'étape en plusieurs problèmes MOO parallèles au niveau de l'instance et dérive efficacement des solutions MOO au niveau de l'étape à partir de solutions MOO au niveau de l'instance. Les résultats de l'évaluation utilisant des charges de travail de production démontrent que notre approche hiérarchique MOO surpasse les méthodes MOO existantes de 4% à 77% en termes de performances et jusqu'à 48% en réduction des coûts tout en fonctionnant dans un délai de 0,02 à 0,23 secondes par rapport aux optimiseurs et planificateurs actuels.Notre troisième contribution vise à atteindre l’optimalité Pareto pour l’ensemble de la requête avec un contrôle plus fin des paramètres. Dans les systèmes Big Data comme Spark, certains paramètres peuvent être ajustés indépendamment pour chaque étape de la requête, tandis que d'autres sont partagés entre toutes les étapes, introduisant ainsi un espace de paramètres de grande dimension et des contraintes complexes. Pour relever ce défi, nous proposons une nouvelle approche appelée MOO hiérarchique avec contraintes (HMOOC). Cette méthode décompose le problème d’optimisation d’un grand espace de paramètres en sous-problèmes plus petits, chacun contraint d’utiliser les mêmes paramètres partagés. Étant donné que ces sous-problèmes ne sont pas indépendants, nous développons des techniques pour générer un ensemble suffisamment large de solutions candidates et les agréger efficacement pour former des solutions Pareto optimales globales. Les résultats de l'évaluation utilisant les benchmarks TPC-H et TPC-DS démontrent que HMOOC surpasse les méthodes MOO existantes, obtenant une amélioration de 4,7% à 54,1% de l'hypervolume et une réduction de 81% à 98,3% du temps de résolution
Sufficiency, consumption patterns and limits: a survey of French households
No full textInternational audienceHow can the concept of consumption corridors be operationalised? This research provides socio-demographic knowledge of the setting of the upper limit. Four distinct ‘modes of consumption’ are identified, based on material consumption levels and openness to consumption limits. A survey of French households (n = 2452) reveals people are generally reluctant to accept strict consumption caps, especially binding ones. Both high and low material consumption groups strongly oppose consumption limits, suggesting that wealth does not correlate with a sense of having ‘enough’. Individuals with fewer possessions support the idea of limits to consumption, though not outright bans. Despite the cultural value placed on limitless consumption and political aversion to restrictions, actual consumption modes are inherently limited. Individuals operate within certain boundaries, whether or not acknowledged. Since perceptions of ‘enough’ are shaped by economic, social and technical contexts, urban settings and buildings could play a critical role in establishing these de facto limits. By facilitating frugal-yet-comfortable lifestyles, cities and buildings could help to restrain consumption without invoking a sense of deprivation. This approach suggests a pathway for fostering sustainable consumption corridors that feel normal rather than imposed. Practice relevance This research identifies four main modes of consumption in relation to sufficiency in mainland France. It shows that what constitutes enough is socially and economically situated and is not an external reality that would mechanically satisfy consumption needs. It also shows a general reluctance of individuals towards setting limits to individual consumption levels. To the extent that urban planning and dwelling types are already important forces in the shaping of sustainable modes of consumption, cities and buildings may prove instrumental in providing the condition of de facto upper limits to consumption. By understanding consumption modes and their relationship with sufficiency, policymakers, urban planners and architects could implement the means to conduct frugal lifestyles that do not evoke feelings of deprivation
Sur la thermomécanique des champs de dislocation
No full textThis thesis investigates the coupling between dislocation evolution and heat conduction in continuum bodies through a theoretical and numerical approach. The main objectives are twofold: (i) to develop a finite deformation theory of thermomechanics of field (i.e. continuously represented) dislocations that account for the interplay between dislocation activity and temperature evolution, while considering only observable fields; (ii) to propose a geometrical linearisation of the finite deformation theory showing that it is similar to the small deformation thermal field dislocation mechanics (TFDM) theory proposed in Upadhyay,J. Mech. Phys. Solids, 145 (2020) 104150, and numerically implement the latter using the finite element (FE) approach to study temperature evolution during dislocation transport.The fundamental aspects of dislocation modelling are reviewed, highlighting the different approaches that have commonly been used to study dislocation-based plasticity in crystals. After identifying the current limitations of the state of the art, a theory with a novel kinematics for thermo-elastoplastic problems based on dislocation mechanics in a finite deformation framework within a transient heterogeneous temperature field is proposed. The theory does not require the specification of a global reference configuration, whence we do not make use of a multiplicative decomposition of the deformation gradient into elastic, plastic, and thermal parts. Instead, considering only observable state variables, we show that the kinematics based on the conservation of Burgers vector is sufficient to yield the commonly-accepted additive decomposition of the velocity gradient into elastic, plastic, and thermal distortion rates. Accounting for the polar dislocation density as a state variable in the Helmholtz free energy of the system, using the first and second laws of thermodynamics, we obtain a new structure of the temperature evolution equation, which allows for solutions in the form of dispersive waves with finite propagation speed without a second derivative of the temperature field in time.The developed theory is shown to reduce, when geometrically linearised, to the small-strain TFDM theory previously proposed. Then, the focus is turned to the latter, and the variational forms of its partial differential equations (PDEs) are presented. Using an open-source library designed to solve PDEs with the FE method, the variational forms are implemented in a staggered algorithm. The implementation is verified against an analytical solution for the temperature field generated by a moving dislocation, and excellent agreement is obtained. Some of the TFDM capabilities are then explored in examples of the heat generated by single edge/screw dislocation, dislocation annihilation, and dislocation loop expansion, which provide a clear understanding of the transient thermoelastic and plastic heat sources involved in each case.The present research advances the field of continuum dislocation modelling by proposing a novel theoretical framework, as well as the numerical implementation of its linearised version. This work serves as a basis for understanding the evolution of dislocation structures during different thermomechanical processes, such as metal additive manufacturing, welding, quenching, etc., which would ultimately contribute to better controlling the mechanical properties of manufactured parts. Future work would include an extension of the numerical implementation to the general finite-deformation theory proposed, as well as an upscaling of the latter to account for the role of statistically stored dislocations in classical problems of plasticity.Cette thèse explore le couplage entre l'évolution des dislocations et la conduction thermique dans les corps continus à travers une approche théorique et numérique. Les principaux objectifs sont : (i) développer une théorie de la thermomécanique des champs de dislocation en grandes déformations qui tient compte de l'interaction mutuelle entre l'activité des dislocations et l'évolution de la température, tout en considérant uniquement des champs observables ; (ii) proposer une linéarisation géométrique de cette théorie en montrant qu'elle revient à la théorie thermomécanique des champs de dislocations (TFDM) en petites déformations proposée par Upadhyay, J. Mech. Phys. Solids, 145 (2020) 104150, et implémenter numériquement cette dernière en utilisant la méthode des éléments finis (EF) pour étudier l'évolution de la température pendant le transport des dislocations.Les aspects fondamentaux de la modélisation des dislocations sont passés en revue, mettant en évidence les différentes approches couramment utilisées. Après avoir identifié les limitations actuelles de l'état de l'art, une théorie avec une nouvelle cinématique basée sur la mécanique des dislocations dans un cadre de grandes déformations considérant un champ de température hétérogène transitoire est proposée. La théorie ne nécessite pas la spécification d'une configuration de référence globale, d'où l'absence de décomposition multiplicative du gradient de déformation en parties élastique, plastique et thermique. Au lieu de cela, en ne considérant que des variables d'état observables, il est montré que la cinématique basée sur la conservation du vecteur de Burgers est suffisante pour obtenir la décomposition additive couramment acceptée du gradient de vitesse en taux de distorsion élastique, plastique et thermique. En prenant en compte la densité de dislocations polaires comme variable d'état dans l'énergie libre de Helmholtz du système, et en utilisant les première et deuxième lois de la thermodynamique, une nouvelle structure de l'équation d'évolution de la température est obtenue, permettant des solutions sous forme d'ondes dispersives avec une vitesse de propagation finie, sans dérivée seconde du champ de température dans le temps.La théorie développée est montrée se réduire, sous linéarisation géométrique, à la théorie TFDM en petites déformations précédemment proposée. Ensuite, l'accent est mis sur cette dernière, et les formes variationnelles de ses équations aux dérivées partielles (EDP) sont présentées. En utilisant une bibliothèque open-source conçue pour résoudre les EDP avec la méthode des EF, les formes variationnelles sont implémentées dans un algorithme échelonné. L'implémentation est vérifiée par rapport à une solution analytique pour le champ de température généré par une dislocation en mouvement, et un excellent accord est obtenu. Certaines des capacités de TFDM sont ensuite explorées dans des exemples de chaleur générée par le mouvement d’une dislocation coin/vis, l'annihilation des dislocations et l'expansion des boucles de dislocations, fournissant une compréhension en profondeur des sources de chaleur thermoélastiques et plastiques transitoires impliquées dans chaque cas.La présente recherche fait progresser le domaine de la modélisation des champs de dislocations en proposant un nouveau cadre théorique, ainsi que l'implémentation numérique de sa version linéarisée. Ce travail sert de base à la compréhension de l'évolution des structures de dislocations lors de différents processus thermomécaniques, tels que la fabrication additive de métaux, le soudage, la trempe, etc., ce qui pourrait contribuer à un meilleur contrôle des propriétés mécaniques des pièces fabriquées. Les travaux futurs incluraient une extension de l'implémentation numérique à la théorie proposée en grandes déformations, ainsi qu'un échelonnement de cette dernière pour tenir compte du rôle des dislocations statistiquement stockées dans les problèmes classiques de plasticité
Versatile Curve Design by Level Set with Quadratic Convergence
No full textInternational audienceMany 3D mesh processing tasks revolve around generating and manipulating curves on surface meshes. While it is intuitive to explicitly model these curves using mesh edges or parametric curves in the ambient space, these methods often suffer from numerical instability or inaccuracy due to the projection operation. Another natural strategy is to adapt spline based tools, these methods are quite fast but are hard to be extended to more versatile constraints and need heavy manual interactions. In this paper, we present an efficient and versatile approach to curve design based on an implicit representation known as the level set. While previous works have explored the use of the level set to generate curves with minimal length, they typically have limitations in accommodating additional conditions for rich and robust control. To address these challenges, we formulate curve editing with constraints like smoothness, interpolation, tangent control, etc., via a level set based variational problem by constraining the values or derivatives of the level set function. However, the widely used gradient flow strategy converges very slowly for this complicated variational problem compared to the classical geodesic one. Thus, we propose to solve it via Newton's method enhanced by local Hessian correction and a trust-region strategy. As a result, our method not only enables versatile control, but also excels in terms of performance due to nearly quadratic convergence and almost linear complexity in each iteration via narrow band acceleration. In practice, these advantages effectively benefit various applications, such as interactive curve manipulation, boundary smoothing for surface segmentation and path planning with obstacles as demonstrated
Characterization of a novel variant in the NR3C1 gene: differentiating glucocorticoid resistance from Cushing Syndrome
No full textInternational audienceAbstract Introduction Primary generalized glucocorticoid resistance syndrome (GGRS) is a rare endocrine disease caused by loss-of-function variants of the NR3C1 gene encoding the Glucocorticoid Receptor. We describe a novel heterozygous missense variant (NM_000176.3, c.1330T>G, p.Phe444Val) within the DNA Binding Domain. Clinical case Elevated urinary-free cortisol levels were detected in a 59-year-old male before bariatric surgery (BMI 39.9 kg/m2). Early-onset hypertension was well controlled. The low dose dexamethasone suppression test was pathologic, but ACTH and midnight salivary cortisol levels were normal. The patient was initially referred to transsphenoidal surgery for a presumed diagnosis of Cushing disease. He presented to our department at the age of 68, when the clinical diagnosis of GGRS was established. Methods Functional characterization of the variant was performed ex vivo through transient transfection assays in HEK 293T cells to assess transcriptional activity and nuclear translocation. Results The variant showed a lack of transcriptional activity (GRWT: 91.5 [80.5; 101.2] vs. GRF444V: 1.0 [1.0; 1.0]) despite efficient nuclear translocation in response to dexamethasone, suggesting a DNA binding defect of the variant. These results are discussed in the light of previously reported GGRS cases. Conclusion We have described a novel heterozygous mutation of the NR3C1 gene associated with primary GGRS. This case highlights the importance of raising awareness of clinical and laboratory features of this rare disorder, to enable early diagnosis and avoid unnecessary and potentially dangerous diagnostic and therapeutic procedures
Fairness and Consensus in an Asynchronous Opinion Model for Social Networks
No full textInternational audienceWe introduce a DeGroot-based model for opinion dynamics in social networks. A community of agents is represented as a weighted directed graph whose edges indicate how much agents influence one another. The model is formalized using labeled transition systems, henceforth called opinion transition systems (OTS), whose states represent the agents' opinions and whose actions are the edges of the influence graph. If a transition labeled (i,j) is performed, agent j updates their opinion taking into account the opinion of agent i and the influence i has over j. We study (convergence to) opinion consensus among the agents of strongly-connected graphs with influence values in the interval (0,1). We show that consensus cannot be guaranteed under the standard strong fairness assumption on transition systems. We derive that consensus is guaranteed under a stronger notion from the literature of concurrent systems; bounded fairness. We argue that bounded-fairness is too strong of a notion for consensus as it almost surely rules out random runs and it is not a constructive liveness property. We introduce a weaker fairness notion, called m-bounded fairness, and show that it guarantees consensus. The new notion includes almost surely all random runs and it is a constructive liveness property. Finally, we consider OTS with dynamic influence and show convergence to consensus holds under m-bounded fairness if the influence changes within a fixed interval. We illustrate OTS with examples and simulations, offering insights into opinion formation under fairness and dynamic influence
urbisphere-Berlin Campaign: Investigating Multiscale Urban Impacts on the Atmospheric Boundary Layer
No full textInternational audienceFor next-generation weather and climate numerical models to resolve cities, both higher spatial resolution and subgrid parameterizations of urban canopy-atmosphere processes are required. The key is to better understand intraurban variability and urban-rural differences in atmospheric boundary layer (ABL) dynamics. This includes upwind-downwind effects due to cities' influences on the atmosphere beyond their boundaries. To address these aspects, a network of >25 ground-based remote sensing sites was designed for the Berlin region (Germany), considering city form, function, and typical weather conditions. This allows investigation of how different urban densities and human activities impact ABL dynamics. As part of the interdisciplinary European Research Council Grant urbisphere, the network was operated from autumn 2021 to autumn 2022. Here, we provide an overview of the scientific aims, campaign setup, and results from 2 days, highlighting multiscale urban impacts on the atmosphere in combination with high-resolution numerical modeling at 100-m grid spacing. During a spring day, the analyses show systematic upwind-city-downwind effects in ABL heights, largely driven by urban-rural differences in surface heat fluxes. During a heatwave day, ABL height is remarkably deep, yet spatial differences in ABL heights are less pronounced due to regionally dry soil conditions, resulting in similar observed surface heat fluxes. Our modeling results provide further insights into ABL characteristics not resolved by the observation network, highlighting synergies between both approaches. Our data and findings will support modeling to help deliver services to a wider community from citizens to those managing health, energy, transport, land use, and other city infrastructure and operations
On the Matchings-Jack and Hypermap-Jack Conjectures for Labelled Matchings and Star Hypermaps
No full textInternational audienceIntroduced by Goulden and Jackson in their 1996 paper, the matchings-Jack conjecture and the hypermap-Jack conjecture (also known as the b-conjecture) are two major open questions relating Jack symmetric functions, the representation theory of the symmetric groups and combinatorial maps. They show that the coefficients in the power sum expansion of some Cauchy sum for Jack symmetric functions and in the logarithm of the same sum interpolate respectively between the structure constants of the class algebra and the double coset algebra of the symmetric group and between the numbers of orientable and locally orientable hypermaps. They further provide some evidence that these two families of coefficients indexed by three partitions of a given integer n and the Jack parameter α are polynomials in β = α -1 with non-negative integer coefficients of combinatorial significance. This paper is devoted to the case when one of the three partitions is equal to (n). We exhibit some polynomial properties of both families of coefficients and prove a variation of the hypermap-Jack conjecture and the matchings-Jack conjecture involving labelled hypermaps and matchings in some important cases.</div