HAL-Polytechnique
Not a member yet
51406 research outputs found
Sort by
Comparison of General Circulation Models of the Venus upper atmosphere
No full textInternational audienceIn the context of future Venusian missions, it is crucial to improve our understanding of Venus upper atmosphere through 3D modeling, notably for spacecraft orbit computation. This study compares three General Circulation Models (GCMs) of the Venusian atmosphere up to the exosphere: the Venus Planetary Climate Model (Venus PCM), the Venus Thermospheric Global Model (VTGCM) and the Tohoku University GCM (TUGCM), focusing on their nominal simulations (e.g. composition, thermal structure and heating/cooling rates). Similarities and discrepancies among them are discussed in this paper, together with data-models comparison. The nominal simulations analyzed in this study fail to accurately reproduce the daytime observations of Pioneer Venus, notably overestimating the exospheric temperature. This is linked to an underestimation of the atomic oxygen (O) abundance in the three GCMs, and suggests the need of additional O production in the thermosphere. The selection of solar spectrum is also the main reason for the discrepancies between the models in terms of temperature dependence on solar activity. A list of recommendations is proposed aiming at improving the modeling of Venus’ upper atmosphere, among them: 1. Standardize the EUV-UV solar spectrum input. 2. Update the near-infrared heating scheme with Venus Express-Era data. 3. Reassess Radiative cooling schemes. 4. Investigate the underestimated atomic Oxygen abundance
Generation of Realistic Geolocated Agendas in a Digital Twin using GPS-based Mobility Dataset
No full textInternational audienceThis work aims to generate individual agendas and locate each individual activity. Probabilistic tools are generated to provide a framework of individual behaviours. Such information could then be injected in a digital twin of community and individuals in order to model how places are occupied and how individuals meet. It uses the well detailed dataset provided by the NetMob 2025 Data Challenge
Byzantine-Robust Gossip: Insights from a Dual Approach
No full textDistributed learning has many computational benefits but is vulnerable to attacks from a subset of devices transmitting incorrect information. This paper investigates Byzantine resilient algorithms in a decentralized setting, where devices communicate directly in a peer-to-peer manner within a communication network. We leverage the so-called dual approach for decentralized optimization and propose a Byzantine-robust algorithm. We provide convergence guarantees in the average consensus subcase, discuss the potential of the dual approach beyond this subcase, and re-interpret existing algorithms using the dual framework. Lastly, we experimentally show the soundness of our method
In vivo autofluorescence lifetime imaging of the Drosophila brain captures metabolic shifts associated with memory formation
No full textInternational audienceAbstract Neuronal energy regulation is increasingly recognized as a critical factor underlying brain functions and their pathological alterations, yet the metabolic dynamics that accompany cognitive processes remain poorly understood. As a label-free and minimally invasive technique, fluorescence lifetime imaging (FLIM) of coenzymes NADH and NADPH (collectively referred to as NAD(P)H) offers the possibility to resolve cellular metabolic profiles with high spatial precision. However, NAD(P)H FLIM’s capacity to detect subtle changes in neuronal metabolism associated with cognition has not been demonstrated. In this study, we applied NAD(P)H FLIM to map the metabolic profiles of Drosophila neurons in vivo across multiple scales, focusing on the primary centers for associative memory: the mushroom bodies (MBs). At a broad scale, we obtained an overview of the metabolic signatures of the main brain tissue and identified a marked difference between neuropil and cortex areas. At a finer scale, our findings revealed notable heterogeneity in the basal metabolic profiles of distinct MB neuron subtypes. Measurements performed after associative olfactory learning also uncovered a subtype-specific metabolic shift associated with memory formation, demonstrating the utility of NAD(P)H FLIM in detecting physiology-driven changes linked to brain function. These results establish a promising framework for studying cerebral energy dynamics in vivo
Nutritional and growth enhancement of alfalfa sprouts through cold plasma and UV seed treatments
No full textInternational audienceEmploying eco-friendly techniques like cold plasma (CP) and ultraviolet (UV) radiation provides innovative approaches to enhance the sprout quality and productivity of alfalfa. This study explores the effects of CP and UV radiation on the germination, growth, and phytochemical profiles of alfalfa sprouts. CP significantly accelerated germination time, reducing median germination time by 8 hours compared to the control, and enhanced photosynthetic pigments, leading to higher biomass (25.87 mg/sprout fresh weight and 1.45 mg/sprout dry weight). UV treatments, particularly UV-C, increased chlorophyll and total flavonoid content. Overall, CP effectively promotes alfalfa germination and growth, while UV treatments improve specific phytochemicals
Nexus approach to enhance water-energy-food security and ecosystems resilience under climate change in the Mediterranean
No full textInternational audienceThe Mediterranean Basin, already a water-scarcity hotspot, faces intensifying droughts and warming that strain the water–energy–food–ecosystems (WEFE) nexus. Climate impacts cascade across sectors, while siloed responses risk maladaptation. Nexus-based solutions—centred on water—can foster synergies and reduce trade-offs, with nature-based, socially inclusive, and clean energy strategies offering transformative potential. Yet governance, cooperation, and data gaps persist; closing these is vital to operationalize the nexus and advance regional sustainability
Interactions et opportunités au croisement entre la modélisation probabiliste par apprentissage profond et l’inférence statistique par Monte Carlo
No full textThis thesis advances the field of sampling, a cornerstone of Bayesian inference, computational physics, and probabilistic modeling, where the goal is to generate samples from a known probability density. A related challenge arises in generative modeling, which seeks to produce new data resembling a given dataset, a problem that has seen major breakthroughs through recent advances in deep learning. The central aim of this work is to leverage modern generative models to enhance classical sampling frameworks. The study begins by examining the inherent difficulties of multi-modal sampling, identifying key limitations of both classical and advanced Monte Carlo methods. It then explores the integration of pre-trained normalizing flows into traditional Monte Carlo schemes, providing practical guidance on their performance across diverse target distributions. Building on this, diffusion models are incorporated into advanced annealed Monte Carlo methods, revealing both their potential and their limitations. The work also investigates how diffusion models can be embedded within a variational inference framework. In parallel, it proposes a learning-free diffusion-based sampler that replaces neural approximators with Monte Carlo estimators. Finally, these enhanced sampling strategies are applied to the training of energy-based models, introducing a novel algorithm in which a normalizing flow serves as an auxiliary sampler to facilitate the training of these expressive yet challenging generative models.Cette thèse fait progresser le domaine de l’échantillonnage, pierre angulaire de l’inférence bayésienne, de la physique computationnelle et de la modélisation probabiliste, dont l’objectif est de générer des échantillons à partir d’une densité de probabilité connue. Un défi connexe apparaît en modélisation générative, qui vise à produire de nouvelles données ressemblant à un jeu de données existant, un problème ayant connu des avancées majeures grâce aux récents progrès de l’apprentissage profond. L’objectif central de ce travail est de tirer parti des modèles génératifs modernes afin d’enrichir les méthodes classiques d’échantillonnage. L’étude commence par examiner les difficultés inhérentes à l’échantillonnage multi-modal, en identifiant les principales limites des méthodes de Monte Carlo classiques et avancées. Elle explore ensuite l’intégration de flows normalisants pré-entraînés dans des méthodes Monte Carlo traditionnelles, en apportant des indications pratiques sur leurs performances pour diverses types de distributions cibles. Dans le prolongement, des modèles de diffusion sont intégrés à des méthodes de Monte Carlo avancées, révélant à la fois leur potentiel et leurs limites. Cette thèse s’intéresse également à l’intégration des modèles de diffusion dans un cadre d’inférence variationnelle. En parallèle, nous proposons un échantillonneur basé sur les modèles de diffusion et exempt d’apprentissage, qui remplace les approximations neuronales usuelles par des estimateurs Monte Carlo. Enfin, ces stratégies d’échantillonnage améliorées sont appliquées à l’entraînement de modèles génératifs basés sur l'énergie, en introduisant un nouvel algorithme où un flow normalisant sert d’échantillonneur auxiliaire pour faciliter l’apprentissage de ces modèles génératifs expressifs mais exigeants
Living on the edge: radius effects in the angular substructure of heavy-ion jets
No full textInternational audienceJet substructure observables serve as essential tools for probing the quark-gluon plasma produced in relativistic heavy-ion collisions. Their interpretation, however, is often complicated by edge effects, which arise when correlated particles fall outside the reconstructed jet radius, introducing distortions that obscure the underlying QCD dynamics. In this work, we present a comprehensive phenomenological study of edge effects in soft-insensitive angular observables, taking the two-point energy correlator (EEC) as a representative example. We argue that these distortions scale linearly with the average angular separation between the winner-take-all and -scheme axes , and validate this behavior across proton-proton (p-p) simulations with Pythia8 and Herwig7, as well as lead-lead (Pb-Pb) simulations using JEWEL and CoLBT. In p-p collisions, edge effects are strongly suppressed, scaling as , whereas medium-modified jets can exhibit larger distortions, with contributions scaling as and . Taking Pb-Pb/p-p ratios of the EEC substantially reduces, but does not completely eliminate, these distortions, highlighting the need of accounting for edge effects in the interpretation of heavy-ion jet substructure measurements. Since edge effects are largely governed by the distribution, studying this distribution provides a new handle for benchmarking and constraining the modeling of edge effects in heavy-ion event generators
Comptabilité Distributionnelle : Contexte, Résultats, Limites et Enjeux
No full textDistributional National Accounts (DINA) result from a reconciliation exercise of tax data, household surveys, and national accounts in order to construct income and wealth series that are fully consistent with the macroeconomic aggregates of the national accounts. This article offers a perspective on this tool. Rather than focusing on methodology and results, the aim is to take a step back from this new statistical object in order to situate DINA within the broader academic movement on long-run inequality, present some of the key findings that emerge from it, and discuss the limitations and future avenue of research of this approach. Three questions guide our analysis: Do DINA represent a methodological break? Are these new measures open to improvement? What new avenues do they offer for research and statistical production?Les comptes nationaux distributionnels (DINA) sont le fruit d'un exercice de réconciliation entre données fiscales, enquêtes auprès des ménages et comptes nationaux afin de construire des séries de revenus et de patrimoine qui soient parfaitement cohérentes avec les agrégats macroéconomiques de la comptabilité nationale. Cet article propose une mise en perspective de cet outil. Plutôt que de se centrer sur la méthodologie et les résultats, l'objectif de cet article est de prendre du recul sur ce nouvel objet statistique pour situer les DINA dans le mouvement académique plus large sur les inégalités de long terme, présenter quelques résultats centraux qui en découlent, et discuter les limites et perspectives de cette approche. Trois questions guideront notre réflexion : les DINA représentent-ils une rupture méthodologique ? Les nouvelles mesures qu'ils proposent sont-elles perfectibles ? Quels horizons ouvrent-ils pour la recherche et la production statistique
Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation
No full textInternational audienceGlobal visual geolocation predicts where an image was captured on Earth. Since images vary in how precisely they can be localized, this task inherently involves a significant degree of ambiguity. However, existing approaches are deterministic and overlook this aspect. In this paper, we aim to close the gap between traditional geolocalization and modern generative methods. We propose the first generative geolocation approach based on diffusion and Riemannian flow matching, where the denoising process operates directly on the Earth's surface. Our model achieves state-of-the-art performance on three visual geolocation benchmarks: OpenStreetView-5M, YFCC-100M, and iNat21. In addition, we introduce the task of probabilistic visual geolocation, where the model predicts a probability distribution over all possible locations instead of a single point. We introduce new metrics and baselines for this task, demonstrating the advantages of our diffusion-based approach. Codes and models will be made available