Archive ouverte de Centrale Lyon
Not a member yet
32420 research outputs found
Sort by
Analysis of spatiotemporal patterns of midwife toads using a three-dimensional array network
No full textInternational audienceDuring the breeding season, male Midwife Toads Alytes obstetricans emit mating calls from land to attract females. From April to August, these calls form choruses of short, tonal sounds. This study aims to describe the spatial and temporal organization of one such chorus near Lyon, France. Initially tested in a semi-anechoic room using playback choruses of green tree frogs, a network of five spherical microphone arrays was deployed in the toads' territory. This work aims, after a geometrical calibration process of the network, to separate and localize the calls within an area of about ten square meters. Detection reports in the temporal domain are then created to compile information on the timing, location, and frequency of specific events. Ultimately, it is possible to have a visual representation of the spatial and temporal distribution of Midwife Toad choruses
Complex formation mechanism between beta-cyclodextrin and organic micropollutants in water. A steered-molecular dynamics simulations study
No full textInternational audienceCyclodextrins (CDs) are cyclic oligosaccharides composed of glucose units, known to form complexes with some organic molecules that are trapped inside the CD cavity. This complexation is widely used in pharmaceutics to improve the solubility of hydrophobic drugs, and is promising in the environmental field to enhance the removal of organic micropollutants from water. The mechanisms underlying the interactions between CDs and guest molecules are driven by hydrophobic effects, electronic and van der Waals interactions, hydrogen bonding, as well as the flexibility of the guest molecules. Nevertheless, the prediction of the driving forces of the complex formation, between CDs and guest molecules, as well as the comprehensive analysis of the binding process, are still challenges.In this study, molecular dynamics (MD) and steered molecular dynamics (SMD) simulations were used to investigate the thermodynamics and the dynamics of the complex formation between b-CDs and three different molecules (tetracycline, trimethoprim, caffeine) in water
Constructions for positional games and applications to domination games
No full textWe present constructions regarding the general behaviour of biased positional games, and amongst others show that the outcome of such a game can differ in an arbitrary way depending on which player starts the game, and that fair biased games can behave highly non-monotonic. We construct a gadget that helps to transfer such results to Maker-Breaker domination games, and by this we extend a recent result by Gledel, Iršič, and Klavžar, regarding the length of such games. Additionally, we introduce Waiter-Client dominations games, give tight results when they are played on trees or cycles, and using our transference gadget we show that in general the length of such games can differ arbitrarily from the length of their Maker-Breaker analogue
Liouville property for groups and conformal dimension
No full textInternational audienceConformal dimension is a fundamental invariant of metric spaces, particularly suited to the study of self-similar spaces, such as spaces with an expanding selfcovering (e.g. Julia sets of complex rational functions). The dynamics of these systems are encoded by the associated iterated monodromy groups, which are examples of contracting self-similar groups. Their amenability is a well-known open question. We show that if G is an iterated monodromy group, and if the (Alfhors-regular) conformal dimension of the underlying space is strictly less than 2, then every symmetric random walk with finite second moment on G has the Liouville property. As a corollary, every such group is amenable. This criterion applies to all examples of contracting groups previously known to be amenable, and to many new ones. In particular, it implies that for every post-critically finite complex rational function f whose Julia set is not the whole sphere, the iterated monodromy group of f is amenable
Transformateur matriciel avec inductance de fuite intégrée
No full textInternational audienceL’électrification croissante du secteur aéronautique impose une amélioration continue des convertisseurs statiques d’électronique de puissance, notamment au niveau de leurs composants magnétiques, qui peuvent représenter plus de 50 % de l’encombrement total du système. Pour répondre aux exigences de compacité, de rendement élevé et de faible coût, l’intégration de l’inductance de fuite au sein du transformateur est une solution prometteuse
Engineering Pb-free relaxor ferroelectric thin films for low voltage energy storage applications
No full textInternational audiencePulsed power technologies demand dielectric capacitors that possess a high energy storage density and efficiency at low applied electric fields/voltages. In this work, we engineered the morphology of lead-free 0.85[0.6Ba(Zr0.2Ti0.8)O3–0.4(Ba0.7Ca0.3)TiO3]–0.15SrTiO3 (BZCT–STO) epitaxial thin films, fabricated using the pulsed laser deposition technique. Through control of the annealing time, we observed both grain shape and size changes, which induced a change in the relaxor behaviour of the BZCT–STO films. The enhanced relaxor behaviour, assigned to the formation of polar nanoregions, was achieved in the film with uniform smaller spherical grains, which is relevant for improved energy storage performance at low electric fields. The dependence of the electric field on the ferroelectric and energy storage properties of the BZCT–STO thin films was investigated. It is found that the LSMO/BZCT–STO/Au capacitor with enhanced relaxor behaviour shows the optimum energy storage performance, attributable to a moderate maximum polarization and remanent polarization difference, and the highest electric breakdown field. An energy storage density of 9.24 J cm−3 with an efficiency of 86.4% at an applied electric field of 1500 kV cm−1 was obtained. The increased energy storage density and efficiency in these BZCT–STO thin film capacitors at a low electric field make them one of the most promising systems reported in the literature for energy storage applications. The results reported here clearly evidence the significant impact of the film morphology on the dielectric, ferroelectric and energy storage properties
Fast automatic multiscale electron tomography for sensitive materials under environmental conditions
No full textfffInternational audienceliste mot clef
Structuration de représentations visuelles pour améliorer la généralisation en apprentissage auto-supervisé
No full textRepresentation learning has become a central pillar of modern artificial intelligence, playing a key role in recent advances in fields such as computer vision and natural language processing. With the growing interest in self-supervised learning, where models independently learn from raw data without human supervision, representation learning has become even more critical. It enables making sense of raw data by extracting relevant features. Furthermore, this autonomous framework promotes the learning of more general representations due to the absence of specific labeling—making them task-agnostic of downstream tasks—while leveraging large quantities of raw data. However, the challenge lies in identifying a supervisory signal, accessible solely from input data, but sufficiently relevant to structure general representations that perform well on downstream tasks. Recent methods for self-supervised visual representation learning employ instance discrimination as a pretext task, demonstrating strong potential to generate rich, reusable, and transferable representations for a wide range of downstream tasks, sometimes even surpassing supervised approaches. The principle of instance discrimination is based on the idea that similar inputs should be projected to similar points in the representation space. In practice, this is typically achieved through a Siamese architecture, which processes two augmented views of the same input using identical networks. These views are generated in a self-supervised manner by applying transformations—also known as augmentations—to the same image, creating pairs that are semantically similar but visually distinct. The pretext task then aims to align the outputs of the two views, encouraging the network to build representations invariant to augmentations, thereby emphasizing the learning of shared visual patterns between the views. This learning approach, grounded in a pretext task designed to capture invariance, differs from historical methods, such as reconstruction-based approaches, which aim to reconstruct an image from its representation. Instance discrimination focuses on a structure-oriented objective, and the success of these approaches highlights the importance of exploring the structural properties of learned representations, not merely as a practical tool for designing pretext tasks but as a direct means of improving their quality. This thesis adopts this perspective, exploring how the structure of representations—notably invariance, sensitivity, and equivariance—can be leveraged to improve generalization in visual representation learning. This issue is addressed through specific sub-questions, each linked to a contribution of the thesis. These sub-questions examine structure through various approaches, such as modifying data distribution, incorporating variational aspects, utilizing equivariance, or analyzing correlations between performance and structural sub-properties. This work has underscored that the structure of representations plays a significant role in generalization and demonstrated that it is an effective lever for improving performance.L'apprentissage de représentations est devenu un pilier central de l'intelligence artificielle moderne, jouant un rôle clé dans les avancées récentes de domaines tels que la vision par ordinateur et le traitement du langage naturel. Avec l'intérêt croissant pour l'apprentissage auto-supervisé, où les modèles apprennent de manière autonome à partir de données brutes sans supervision humaine, l'apprentissage des représentations est devenu encore plus important. Il permet de donner un sens à ces données brutes en en extrayant des caractéristiques pertinentes. De plus, ce cadre autonome favorise l'apprentissage de représentations plus générales grâce à l'absence d'une labellisation spécifique — ce qui les rend agnostiques aux tâches aval — tout en tirant parti des grandes quantités de données brutes disponibles. Néanmoins, la difficulté réside dans la recherche d'un signal de supervision, accessible uniquement à partir des données d'entrée, mais suffisamment pertinent pour structurer des représentations générales offrant de bonnes performances sur les tâches aval. Les méthodes récentes d'apprentissage auto-supervisé de représentations visuelles utilisent comme supervision des tâches prétextes de discrimination d'instances, qui ont démontré un fort potentiel pour générer des représentations riches, réutilisables et transférables à un large éventail de tâches aval, surpassant parfois même les approches supervisées. Le principe de discrimination d'instances repose sur l'idée que des entrées similaires doivent être projetées vers des points similaires dans l’espace des représentations. En pratique, cela est généralement réalisé grâce à une architecture siamoise, qui traite deux vues augmentées d’une même entrée à travers des réseaux identiques. Ces vues sont générées de manière auto-supervisée en appliquant des transformations — aussi appelées augmentations — sur une même image, produisant des paires sémantiquement similaires mais visuellement distinctes. La tâche prétexte vise ensuite à aligner les sorties des deux vues, encourageant le réseau à construire des représentations invariantes aux augmentations, mettant ainsi l’accent sur l'apprentissage des motifs visuels partagés entre les vues. Cet apprentissage, fondé sur une tâche prétexte visant à capturer une invariance, se distingue des méthodes historiques, telles que celles basées sur la reconstruction, qui cherchent à reconstruire une image à partir de sa représentation. En effet, la discrimination d'instances se focalise sur un objectif orienté structure, et le succès de ces approches met en évidence l'importance d'explorer les propriétés structurelles des représentations apprises, non pas uniquement comme un outil pratique pour concevoir des tâches prétextes, mais comme une façon directe pour améliorer leur qualité. Cette thèse s'inscrit dans cette perspective en explorant comment la structure des représentations — notamment l'invariance, la sensibilité et l'équivariance — peut être exploitée pour améliorer la généralisation dans l'apprentissage des représentations visuelles. Cette problématique est abordée à travers des sous-questions spécifiques, chacune liée à une contribution de la thèse. Ces sous-questions examinent la structure via divers moyens, tels que la modification de la distribution des données, l'ajout d'aspects variationnels, l'utilisation de l'équivariance, ou encore la corrélations entre performances et sous-propriétés structurelles. Ces travaux ont ainsi permis de mettre en lumière que la structure des représentations joue un rôle important dans la généralisation et montrent donc qu'elle constitue donc un levier efficace pour améliorer les performances
A random polymer approach to the weak disorder phase of the vertex reinforced jump process
No full textIn this paper, we study the transient phase of the Vertex Reinforced Jump Process (VRJP) in dimension d ≥ 3. In [29], the authors introduce a positive martingale and show that the VRJP is recurrent if and only if that martingale converges to 0. On , d ≥ 3, with constant conductances W , it can be shown that there is a critical value 0 < Wc() < ∞, such that the martingale converges to 0 if W < Wc() or to a positive limit if W > Wc(). On the other hand, the VRJP martingale can be interpreted as the partition function of a non-directed polymer with a very specific 1-dependent random potential. In this paper, we focus on the question of the Lp integrability of the VRJP martingale, which is related to the (diffusive) behavior of the VRJP. First, taking inspiration from the work of Junk [15] for directed polymers in , we prove that on the half-space of , for all W > Wc() there is some δ > 0 such that the VRJP martingale is in L. Second, we prove that, in dimension d ≥ 4, the VRJP martingale is in L for all p > 1 above the slab critical point W) = limm→∞ Wc( x [-m, m] We also propose some related conjectures
Vers une évaluation temps-réel des capacités de détection d’événements infrasons via l'estimation des pertes de transmission par méthodes d''apprentissage profond
No full textAccurate modeling of infrasound transmission loss is essential for evaluating the performance of the International Monitoring System, enabling the effective design and maintenance of infrasound stations to support compliance of the Comprehensive Nuclear-Test-Ban Treaty. State-of-the-art propagation modeling tools enable transmission loss to be finely simulated using atmospheric models. However, the computational cost prohibits the exploration of a large parameter space in operational monitoring applications. To address this, recent studies made use of a deep learning algorithm capable of making transmission loss predictions almost instantaneously. However, the use of nudged atmospheric models leads to an incomplete representation of the medium, and the absence of temperature as an input makes the algorithm incompatible with long range propagation. In this study, we address these limitations by using both wind and temperature fields as inputs to a neural network, simulated up to 130 km altitude and 4, 000 km distance. We also optimize several aspects of the neural network architecture. We exploit convolutional and recurrent layers to capture spatially and range-dependent features embedded in realistic atmospheric models, improving the overall performance. The neural network reaches an average error of 4 dB compared to full parabolic equation simulations and provides epistemic and data-related uncertainty estimates. Its evaluation on the 2022 Hunga Tonga-Hunga Ha’apai volcanic eruption demonstrates its prediction capability using atmospheric conditions and frequencies not included in the training. This represents a significant step towards near real-time assessment of International Monitoring System detection thresh olds of explosive sources.* We provide a convolutional recurrent neural network estimating in near real-time ground-level infrasound transmission loss.* The neural network exploits spatial and range-dependent features in atmospheric models, and makes predictions with an uncertainty estimate.* The network can be used as a tool for near real-time estimation of infrasound event detection capability at a global scale.* Nous proposons un réseau de neurones récurrent convolutif qui permet d'estimer en temps quasi-réel les pertes par transmission des infrasons au niveau du sol.* Le réseau de neurones exploite les caractéristiques spatiales et dépendante de la distance des modèles atmosphériques d'entrée, et fait des prédictions avec une estimation de l'incertitude.* Le réseau peut être utilisé comme outil pour l'estimation en temps quasi-réel de la capacité de détection d’événements infrasons à l'échelle mondiale