HAL Université des Antilles
Not a member yet
25497 research outputs found
Sort by
Energy consumption for hydrogen production in a single-cell hydrogen sulfide electrolysis system
No full textInternational audienc
Towards Participatory Monitoring of Global Changes at a Local Scale: Challenging the Contribution of Living Labs
No full textInternational audienceAssessing the local scale effects of global changes and share related knowledge with the population may bea key to rural society robustness in the Anthropocene. In line with transdisciplinarity trend, one approachaims to refine knowledge of global change impacts by integrating vernacular knowledge and local observa-tions. However, bridging the gap between theory and practice remains a significant challenge. Living labsare emerging in many regions as a science-society dialogue tool, enabling participatory data collection at alocal scale and analysis that combines scientific knowledge with vernacular expertise.We present experiments from a living-lab in the Cévennes, a Mediterranean mountain rural region in southFrance. Its climate is characterised by summer droughts and a high precipitation variability. Small hy-draulic heritage punctuates the watersheds, reflecting the populations’ adaptation to this variability and thedevelopment of deep vernacular hydrological knowledge. Nowadays, climate change combined to societalmutations (mains water usages and tourism development especially) induces an increasing water scarcity.The region has experimented water scarcity crises in 2017, 2021 and 2022, making hydrology a major localconcern. In this context, the living lab fosters the creation and animation of a inhabitants/scientists coop-eration to propose and experiment water management adapted to the local context.At the current state of this living-lab, the co-collection of data is effective through participatory mappingand participative observation procedure of water flows and biochemistry. This could lead to high resolutiondatabase. However, many questions remain regarding the participatory monitoring that could guaranteelong-term monitoring of relevant and reliable indicators of local-scale effects of global change: Are scientifi-cally relevant indicators the same as those of interest to local inhabitants? How can they be engaged overtime? What measurement tools should be implemented? How should this data be shared? Based on theexperience of this living lab, we will present our results and questioning about the participatory approachesfor global changes monitoring at a local scale and about the inhabitants/critical-zone scientists cooperation
Tessa Alexander, une carte matriarcale dans une approche décentrée
No full textInternational audienceLe temps, comme le rappellent les axes possibles de réflexion émis dans l’appel à communication, est « une donnée fondamentale de l’existence humaine, une notion importante en esthétique dans le domaine artistique. Le temps qui nous inscrit dans une durée de vie plus ou moins longue, mais de toute manière limitée, ne cesse de nous tourmenter ». En reconstituant une carte matrilinéaire, l’artiste trinidadienne Tessa Alexander déploie son ascendance maternelle, s’affilie, recherche le jardin, presque secret, de sa mère. Dans cette œuvre intitulée « Mapping Convergence », se côtoient des documents, des archives iconographiques, teintée par une maitrise technique de l’aquarelle, rehaussement des contours et des emprunts aux cartes postales anciennes de la Caraïbe. L’artiste place de ce jardin maternel à l’épreuve du temps, des temps, des époques qui ont bâtie son île natale. Elle y esquisse une identité faite de rhizome portée par ses femmes, ses mères d’époques différentes, d’origine différente mais qui toujours mettent en exergue un amour leur terre et les cultures qui s’y déploient. Mais à travers cette œuvre les documents d’archive deviennent atemporelle, ils continuent d’exister, de survivre à leur auteur. L’œuvre leur offre une nouvelle vie. Cette œuvre s’inscrit dans le temps d’aujourd’hui où les artistes questionnent la décolonisation des savoirs, déconstruisent les schèmes de l’histoire. Nous questionnerons l’impact des écrits de Gayatri Spivak et de Patricia Mohammed dans la création de cette œuvre
Apprentissage profond et approches computationnelles pour l'analyse, la prédiction et la génération de protéines
No full textProteins, macromolecules essential for life processes, adopt complex three-dimensional structures that determine their functions. Structural bioinformatics aims to develop computational tools to understand sequence-structure-function relationships.This thesis falls within this research area and proposes a multi-scale approach, from local to global, for the analysis, prediction, and generation of protein structures, combining novel computational approaches and deep learning methods. The first part of this work presents a series of contributions to structural analysis : PYTHIA, a deep learning predictor of local conformations; SWORD2 and ICARUS, tools for hierarchical domain decomposition and flexible structural alignment; and PEGASUS, a method for predicting protein flexibility from sequence. These tools form a cohesive software suite for studying proteins at different organizational scales. The second part of this thesis explores new perspectives made possible by the considerable growth of genomic data and the recent emergence of protein language models (pLMs), notably the interpretation of very high-dimensional vector representations (embeddings) derived from pLMs. Among other developments, we propose an Adversarial Autoencoder (AAE) architecture to compress these variable-sized representations into a fixed-dimensional vector, forming a continuous and semantically rich latent space. The validation of this approach is carried out through several downstream predictions of structural and functional properties. Despite compressing these embeddings by a factor of 200, the reconstructed embeddings maintain performance equivalent to that of the original representations for predicting SSP3 and SSP8 secondary structures, among others. Furthermore, after reconfiguring the latent space via a triplet network, they significantly improve structural fold recognition, surpassing structure-based methods. Finally, we explore the generative potential of this latent space. Preliminary interpolation experiments in this space have confirmed its continuity and its potential for generating new protein sequences that fold into plausible structures. All of these works provide a suite of approaches for protein analysis and establish a robust framework for generative modeling.Les protéines, macromolécules essentielles aux processus nécessaires à la vie, adoptent des structures tridimensionnelles complexes qui déterminent leurs fonctions. La bioinformatique structurale a pour objectif de développer des outils computationnels afin de comprendre les relations séquence-structure-fonction. Cette thèse s'inscrit dans ce domaine de recherche et propose une approche multi-échelle, du local au global, pour l'analyse, la prédiction et la génération de structures protéiques, en combinant de nouvelles approches computationnelles et des méthodes d'apprentissage profond. La première partie de ce travail présente une série de contributions à l'analyse structurale : PYTHIA, un prédicteur de conformations locales par apprentissage profond ; SWORD2 et ICARUS, des outils pour la décomposition hiérarchique en domaines et l'alignement structural flexible ; et PEGASUS, une méthode de prédiction de la flexibilité protéique à partir de la séquence. Ces outils forment une suite logicielle cohérente pour l'étude des protéines à différentes échelles d'organisation. La seconde partie de cette thèse s'intéresse aux nouvelles perspectives rendues possibles par la croissance considérable des données génomiques et l'émergence récente des modèles de langage protéique (pLMs), notamment l'interprétation de représentations vectorielles (embeddings) de très haute dimension issues de pLMs. Nous y développons entre autres une architecture d'Auto-Encodeur Adversarial (AAE) pour compresser ces représentations de taille variable en un vecteur de dimension fixe, formant un espace latent continu et sémantiquement riche. La validation de cette approche est réalisée à travers plusieurs prédictions de propriétés structurales et fonctionnelles en aval. Malgré la compression de ces embeddings d'un facteur 200, les embeddings reconstruits conservent des performances équivalentes à celles des représentations originales pour la prédiction de structures secondaires SSP3 et SSP8 entre autres. De plus, après une reconfiguration de l'espace latent via un réseau triplet, ils améliorent significativement la reconnaissance de repliements structuraux (fold) en surpassant les méthodes basées sur la structure. Enfin, nous explorons le potentiel générateur de cet espace latent. Des expériences préliminaires d'interpolation dans cet espace ont confirmé sa continuité et son potentiel pour la génération de nouvelles séquences protéiques se repliant en structures plausibles. L'ensemble de ces travaux fournit une suite d'approches pour l'analyse des protéines et établit un cadre robuste pour la modélisation générative
Genomic evidence of a complex supergene system linking dispersal to social polymorphism
No full textInternational audienceSupergenes play an important role in the evolution of complex phenotypes, locking specific combinations of alleles 1-3 . One such complex phenotypes is social polymorphism in ants, i.e., the presence of single (monogyne) or multiple (polygyne) reproductive queens within colonies, known to be determined by social supergenes in at least five lineages 4,5 . This polymorphism is associated with divergent individual and colonial traits, some linked to dispersal 6 . We explored the idea that antagonistic selection between social forms favours the emergence of regions of suppressed recombination. To this end, we studied the ant Myrmecina graminicola, in which a social polymorphism is coupled with the presence/absence of wings in queens. We discovered a new "social supergene" of ~20 Mb with three haplotypes. Supergene genotypes determine the three queen phenotypes observed in nature: monogyne winged, monogyne apterous and polygyne apterous. The two haplotypes associated with aptery carry an additional copy of a gene probably involved in wing development as part of a ~116 kb insertion predating the origin of the social supergene (~20Mya vs. ~1Mya). Syntenic analyses showcased an independent evolution of the social supergene. The screening of workers' genotypes suggests that assortative mating and segregation distortion may play a role in preserving supergene polymorphism. This unique supergene system illustrates the theoretically expected genetic link between social polymorphism and dispersal in ants. Its modular evolution mirrors the role of sexually antagonistic selection in the origin of sex chromosomes and makes ants a very promising model for studying supergene evolution.</div
Energy Balance on LEACH-C Based Protocol for Wireless Sensor Network
No full textInternational audienceThis manuscript presents an improvement of the low energy adaptive clustering hierarchy centralized protocol using the energy balance formulation, referred to as the Low Energy-Variance Adaptive clustering (LEVA) protocol. Cluster heads (CHs) vary dynamically according to the remaining energy variation of the overall network, the number of alive wireless sensor devices, and their relative energy dispersion levels. In addition, the existence of CHs is guaranteed in each round. The energy balance formulation helps remove all additional communication between the base station and all WSDs. Thus, LEACH-C becomes a decentralized protocol. The simulation results demonstrate the feasibility and effectiveness of the LEVA protocol in terms of energy balance and network lifetime. The lifetime is increased by 43%, 14%, and 12.5% compared to LEACH, LEACH-C, and LEACH-WSDN protocols, respectively.</div
Energy Quantification of Machine learning for Dynamic Spectrum Access in LoRaWAN Device
No full textInternational audienceLow-power wide-area networks (LPWANs), such as LoRaWAN, must balance device lifetime with an increasingly congested sub-GHz spectrum. Machine learning (ML) approaches, such as the Upper Confidence Bound (UCB) multi-armed bandit algorithm, allow for collision-aware channel selection without gateway-level coordination. However, these approaches' computational and memory requirements could adversely affect device autonomy. This study presents an analysis of the power consumption of the UCB approach on an STM32WL55 LoRaWAN system on a chip (SoC). Current traces were captured for payloads ranging from 10 to 50 bytes and transmission powers ranging from 8 to 16 dBm. The standard algorithm was compared to a UCB algorithm. The results show that the UCB algorithm's power consumption is less than 1% per uplink-ACK cycle: 0.60% during transmission, and 0.70% and 0.30% during the Rx1 and Rx2 windows, respectively. This overhead becomes negligible as payload size or transmission power increases. These results confirm that lightweight learning is a practical solution for dynamic spectrum access in large-scale IoT deployments.</div
Intranasal abuse of buspirone: Cases serie
No full textInternational audienceBuspirone is a non-benzodiazepine anxiolytic from the azapirone class that acts primarily on the serotonergic system. Its mechanism of action is based on partial agonism at 5-hydroxytryptamine 1A receptor (5-HT1A) receptors, acting at both presynaptic and postsynaptic sites. It also exhibits moderate antagonism at presynaptic dopamine D2receptors, as well as D3 and D4, contributing to its distinct pharmacological profile.Buspirone was first approved in 1986 by the FDA:Food and Drug Administration (FDA) and has been used to treat anxiety disorders, such as generalized anxiety disorder, and relieve symptoms of anxiety.We describe here a series of five cases of buspirone abuse spontaneously reported to the addictovigilance center of Montpellier since 2023
