HAL Portal Univ-Côtedazur
Not a member yet
95391 research outputs found
Sort by
Résolution du Voyageur de Commerce à l'aide du Positional Encoding
No full textWe propose transformer-based neural solvers for the Euclidean Traveling Salesman Problem (TSP) that rely on positional encodings rather than coordinate projections. By adapting ALiBi and RoPE, modern positional encodings originally developed for large language models, to the Euclidean setting, our Positional Encoding-based Neural Solvers (PENS) inherit useful invariances and locality biases. To address the increased density of large instances, we introduce a simple yet effective rescaling of city coordinates that further boosts performance. Trained only on TSP-100, PENS achieves state-of-the-art results for instances with up to 10 000 cities, a scale that was previously dominated by methods requiring graph sparsification. These findings demonstrate that positional encodings provide effective inductive biases for neural combinatorial optimization
Conception d'un système focalisant pour les communications point-à-point en bande millimétrique
No full textAs global wireless data demand continues to increase, more than one billion households around the world still lack access to reliable broadband connectivity. Fixed Wireless Access (FWA), especially when operating in millimeter-wave frequency bands, presents a compelling alternative to fiber and copper in underserved areas. However, millimeter-wave FWA faces considerable technical challenges, including high free-space path loss, limited beam-steering agility, and increased RF power consumption due to large phased array architectures. This thesis investigates a compact, high-gain antenna system that combines a 8 × 8 phased array with a Bragg-based dielectric reflectarray. This hybrid configuration enables efficient beam steering over a ± 15° angular range at 27 GHz, while requiring only a limited number of active RF chains. For operation at 39 GHz, a conventional circularly polarized reflectarray is used, positioned behind the Bragg-based reflectarray. Thanks to its engineered bandgap characteristics, the Bragg reflectarray becomes transparent at this frequency, enabling dual-band operation within a single radiating aperture.By selectively activating only a 2 x 2 sub-array for each beam, the system reduces power consumption and thermal load compared to traditional fully active phased arrays. Moreover, polarization diversity is integrated to strengthen link robustness and mitigate path loss effects in non-line-of-sight and multipath environments. The proposed antenna system has been validated through both simulation and experimental measurements. The results demonstrate that the proposed system meets the requirements of next-generation millimeter-wave FWA applications, where energy efficiency, spectral versatility, and compactness are critical design criteria.Alors que la demande mondiale en données sans fil ne cesse de croître, plus d'un milliard de foyers dans le monde restent privés d'un accès fiable à une connexion haut débit. L'Accès Fixe Sans Fil (FWA), notamment dans les bandes de fréquences millimétriques, constitue une alternative pertinente aux réseaux filaires (fibre optique ou cuivre) dans les zones mal desservies. Toutefois, le FWA en bande millimétrique se heurte à plusieurs défis techniques importants, tels que de fortes pertes de propagation en espace libre, une agilité limitée du dépointage de faisceau, ainsi qu'une consommation énergétique accrue liée aux architectures à réseaux phasés de grande taille. Cette thèse propose un système d'antenne compact et à gain élevé, combinant un réseau phasé 8 × 8 avec un réseau-réflecteur diélectrique à miroir de Bragg. Cette architecture hybride permet un balayage efficace du faisceau sur un secteur angulaire de ± 15° à 27 GHz, tout en ne nécessitant qu'un nombre limité de chaînes RF actives. À 39 GHz, un réseau-réflecteur conventionnel à polarisation circulaire est utilisé, positionné derrière le réseau-réflecteur à miroir de Bragg. Grâce à ses propriétés de bande interdite, celui-ci devient transparent à cette fréquence, permettant un fonctionnement bi-bande avec une seule ouverture rayonnante. En activant uniquement un sous-réseau 2 × 2 par faisceau, le système permet de réduire considérablement la consommation de puissance et la charge thermique par rapport aux réseaux phasés entièrement actifs. Par ailleurs, l'intégration d'une diversité de polarisation améliore la robustesse des liaisons et atténue les pertes en environnements complexes ou sans visibilité directe. Le système proposé a été validé par des simulations et des mesures expérimentales, confirmant sa pertinence pour les futures applications FWA en bande millimétrique, où efficacité énergétique, flexibilité spectrale et compacité sont des critères déterminants
CHIL: THE HIDDEN ARCHITECT OF THE ISOFLAVONOID METABOLON?
No full textInternational audiencePlants synthesize a wide variety of molecules essential for their development and adaptation to the environment, as well as bioactive compounds exploited by humans. To this end, metabolic pathways are organized into dynamic multi-enzyme complexes, called metabolons, which facilitate the creation and regulation of specific metabolic networks. However, the modalities of association and organization of the partners of these metabolons remain poorly elucidated, and their impact on the overall metabolism of the plant remains to be clarified. In addition, the question of the possible implication of non-catalytic proteins in these complexes has recently arisen, without their role being clearly defined.In this project, we focused on the study of the structure and function of a metabolon involved in the production of genistein, synthesized by soybean (Glycine max). Genistein is of major interest because of its potential application in estrogen-dependent cancer chemotherapy, and its role in soybean fitness. Here we present our investigations on the interactions between pairs of proteins, with a particular focus on those involving CHIL. Our experimental results, based for instance on BiFC and Size Exchange Chromatography protocols, highlighted the formation of homodimers (CHS, CHI) or trimers (CHIL) when each enzyme is considered individually but also shed light on specific stable enzyme-CHIL associations, such as CHS-CHIL, IFS-CHIL, HID-CHIL…Molecular modeling studies have provided insights into the structural factors underlying the stabilization of these complexes. Long molecular dynamics simulations, together with estimation of G of bindings confirmed the formation of stable CHS dimers, but also showed that CHIs and CHIL are expecting to behave as monomers in the metabolon. Moreover, interesting structures of complexes involving CHIL exhibit binding conformations consistent with substrate channeling events of metabolites between active sites
Syntactic study of self-repair and self-reformulation in French and Spanish: Effects of utterance length
No full textInternational audienceIn this article we propose a study of self-repair and self-reformulation based on the analysis of a corpus of nearly 19,000 tokens, fully segmented into utterances. This corpus includes two languages, French and Spanish, which are equally represented, and five different communicative contexts (sociolinguistic interview, public interview, work meeting, meeting between friends and service interaction) for each of these two languages. This corpus architecture, as well as the segmentation of the entire corpus into utterances, allows us to point out some general trends, not previously described, regarding the effect of unit length on self-repair. Indeed, the average length in tokens of the utterance not only predicts the proportion of units with at least one self-repair, but also models the site of initiation of self-repair and type: if less than 10 % of utterances with 5 tokens or less have self-repair, almost 40 % of utterances with between 11 and 20 tokens have self-repair, and about 70 % of utterances with more than 31 tokens have self-repair. Moreover, in utterances with 5 tokens or less, self-repair is most likely to be without reformulation and to occur in a unit without a predicate, whereas in utterances with more than 11 tokens self-repair is likely to be with or without reformulation and to be initiated after the predicate, towards the middle of the unit
Le droit parlementaire vu par... les constitutionnalistes
No full textConférence organisée par l'UMR DICE Aix-en-Provence université dans le cadre des "Déjeuners du droit parlementaire", Séance 1, sous la dir. de Damien Connil, Priscilla Jensel-Monge et Audrey de Monti
Interactive Optimization of Scaffolded Procedural Patterns
No full textInternational audienceA procedural program is the representation of a family of assets that share the same structural or semantic properties, whose final appearance is determined by different parameter assignments. Identifying the parameter values that define a desired asset is usually a time-consuming operation, since it requires manually tuning parameters separately and in a non-intuitive manner. In the domain of procedural patterns, recent works focused on estimating parameter values to match a target render or sketch, using parameter optimization or inference via neural networks. However, these approaches are neither fast enough for interactive design nor precise enough to give direct control. In this work, we propose an interactive method for procedural parameter estimation based on the idea of scaffolded procedural patterns. A scaffolded procedural pattern is a sequence of procedural programs that model a pattern in a coarse-to-fine manner, in which the desired pattern appearance is reached step-by-step by inheriting previously optimized parameters. Through scaffolding, patterns are more straightforward to sketch for users and easier to optimize for most algorithms. In our implementation, patterns are represented as procedural signed distance functions whose parameters are estimated with a gradient-free optimization method that runs in real-time on the GPU. We show that scaffolded patterns can be created with a node-based interface familiar to artists. We validate our approach by creating and interactively editing several scaffolded patterns. We show the effectiveness of scaffolding through a user study, where scaffolding enhances both the output quality and the editing experience with respect to approaches that optimize the procedural parameters all at once. We also perform a comparison with previous strategies and provide several recordings of real-time editing sessions in the accompanying materials
Passive Acoustic Monitoring of fish with Distributed Acoustic Sensing
No full textUnderwater soundscapes are critical indicators of marine ecosystem health, but traditional passive acoustic monitoring (PAM) with hydrophones is spatially limited and logistically challenging for long-term observations. This study investigates the potential of distributed acoustic sensing (DAS) to overcome these limitations. A field trial was conducted using a bespoke fiber-optic cable deployed off the coast of Monaco, focusing on three sound-producing Mediterranean species: Ophidion rochei, Sciaena umbra, and Epinephelus marginatus, all known for their distinct acoustic signatures during reproductive activity. DAS recordings reveal species-specific patterns consistent with documented behaviors, and validated by simultaneous hydrophone recording. Thanks to its high sensor density, DAS enables the isolation and localization of individual sounds over tens of meters. The results demonstrate that DAS can reliably detect and locate fish sounds under real-world conditions, while also assessing ambient noise levels. If the DAS setup of this experiment may not match the performance of conventional hydrophones at frequencies above 1kHz, the many other advantages of the approach point to a high potential for marine conservation in the context of climate change and biodiversity loss
Semantically Enriched Datasets for Link Prediction: DB100k+, NELL-995+ and YAGO3-10+
No full textInternational audienceKnowledge graphs constitute a native neuro-symbolic experimental setting due to their logic foundations, which motivates the development of neuro-symbolic approaches for Link Prediction (LP). Since current LP reference datasets seldom involves ontological knowledge, benchmarking such approaches is difficult. That is why, starting from the widely accepted datasets DB100k, NELL-995 and YAGO3-10, we semantically enriched them with ontological knowledge, namely class hierarchy and relation signatures (domains and ranges), and inferred new entity type assertions to create DB100k+, NELL-995+ and YAGO3-10+. We also present a generic masking script to generate sub-graphs with variable proportions of triples with signed/partially signed (no domain or no range)/unsigned (no domain and no range) relations, to evaluate the impact of semantic information on learning performance.</div
Hybrid model to simulate optical systems combining metasurfaces and classical refractive elements
No full textThis paper presents an accurate and fast numerical method to compute imaging performance of a centimeter-scale optical system comprising at least one metasurface that can be used inside an optimization design loop to allow direct design of the full system. In order to fit in the optimization design process, the optical response of the metasurfaces is approximated with a surrogate model of the locally periodic response of their constitutive meta-atoms described by a relevant set of variables. This surrogate model is used to obtain performances of imaging systems containing both metasurfaces and refractive lenses using either the generalized Snell's law or a more physically accurate hybrid ray optics/wave optics model that accounts for the diffraction occurring at metasurfaces. We show that the hybrid model is necessary to obtain an accurate prediction of the Modulus Transfer Function of systems containing metasurfaces with non-negligible optical power
ProMM-RS: Exploring Probabilistic Learning for Multi-Modal Remote Sensing Image Representations
No full textInternational audienceRemote sensing imagery offers diverse modalities, such as synthetic aperture radar and multispectral data, which can bring rich, complementary and valuable information about observed scenes. This information is of paramount importance for downstream applications (e.g. land cover mapping, natural resources monitoring, human settlement characterization) that may benefit from such complementarity. Remote sensing imagery often suffers from a lack of labeled data which can hamper the learning of good representations via state-of-the-art supervised methods. Self-supervised learning has thus emerged as a promising paradigm for remote sensing feature extraction, enabling the extraction of meaningful features without reliance on labeled data. While existing multi-modal contrastive models effectively capture shared information between modalities, they often struggle to account for the inherent heterogeneity of multi-modal remote sensing data. This limitation prevents them from fully leveraging the complementarity of multi-modal remote sensing data. Probabilistic representation learning has emerged as a powerful approach to capture the inherent uncertainty and diversity in multi-modal relationships. In this paper we present ProMM-RS, a novel multi-modal self-supervised training framework incorporating a joint probabilistic embedding space to explicitly model the uncertainty of representations between different inputs and modalities. We evaluate our learned representations with a scene classification downstream task from Sentinel optical and radar images, effectively showing the potential of probabilistic embeddings as a way to measure the relevancy of each modality representation, especially under an obstructed dataset