HAL Collection UNC (Univ. de la Nouvelle Calédonie)
Not a member yet
7579 research outputs found
Sort by
Injure raciale : analyse des éléments extrinsèques, au-delà du seul contexte: Crim. 25 févr. 2025, n<sup>o</sup> 24-80.941
No full textInternational audienceSi le fait de réduire une personne à son origine supposée ne présente pas, à lui seul, un caractère injurieux, les juges du fond doivent apprécier le sens et la portée des propos poursuivis en procédant à une analyse des termes du discours dans lequel ils se sont insérés, la seule référence au contexte local étant insuffisante
TACIT : un outil d’évaluation et d’apprentissage adaptatif pour l’enseignement de la compréhension écrite et du vocabulaire
No full textInternational audienc
Multimodal Learning with Uncertainty Quantification based on Discounted Belief Fusion
No full textInternational audienceMultimodal AI models are increasingly used in fields like healthcare, finance, and autonomous driving, where information is drawn from multiple sources or modalities such as images, texts, audios, videos. However, effectively managing uncertainty-arising from noise, insufficient evidence, or conflicts between modalities-is crucial for reliable decision-making. Current uncertainty-aware ML methods leveraging, for example, evidence averaging, or evidence accumulation underestimate uncertainties in high-conflict scenarios. Moreover, the stateof-the-art evidence averaging strategy struggles with non-associativity and fails to scale to multiple modalities. To address these challenges, we propose a novel multimodal learning method with order-invariant evidence fusion and introduce a conflict-based discounting mechanism that reallocates uncertain mass when unreliable modalities are detected. We provide both theoretical analysis and experimental validation, demonstrating that unlike the previous work, the proposed approach effectively distinguishes between conflicting and non-conflicting samples based on the provided uncertainty estimates, and outperforms the previous models in uncertainty-based conflict detection.</div
Spatio-temporal risk prediction of leptospirosis: A machine-learning-based approach
No full textInternational audienceBackground: Leptospirosis is a neglected zoonotic disease prevalent worldwide, particularly in tropical regions experiencing frequent rainfall and severe cyclones, which are further aggravated by climate change. This bacterial zoonosis, caused by the Leptospira genus, can be transmitted through contaminated water and soil. The Pacific islands bear a high burden of leptospirosis, making it crucial to identify key factors influencing its distribution. Understanding these factors is vital for developing targeted policy decisions to mitigate the spread of Leptospira.Methodology/Principal findings: This study aims to establish a precise spatio-temporal risk map of leptospirosis at a national scale, using binarized incidence rates as the variable to predict. The spatial analysis was conducted at a finer resolution than the city level, while the temporal analysis was performed on a monthly basis from 2011 to 2022. Our approach utilized a comprehensive strategy combining machine learning models trained on binarized incidences, along with descriptive techniques for identifying key factors. The analysis encompasses a broad spectrum of variables, including meteorological, topographic, and socio-demographic factors. The strategy achieved a concordance metric of 83.29%, indicating a strong ability to predict the presence of contamination risk, with a sensitivity of 83.93%. Key findings included the identification of seasonal patterns, such as the impact of the El Niño Southern Oscillation, and the determination that rainfall and humidity with a one-month lag are significant contributors to Leptospira contamination. Conversely, soil types rich in organic matter may reduce bacterial presence and survival.Conclusions/Significance: The study highlights the significant influence of environmental factors on the seasonal spread of Leptospira , particularly in tropical and subtropical regions. These findings are crucial for public health planning, providing insights for targeted policies to reduce leptospirosis, while advanced machine learning models serve as a robust tool for improving disease surveillance, and risk assessment, which ultimately supports the development of an early warning system
Spatial autocorrelation and host anemone species drive variation in local components of fitness in a wild clownfish population
No full textInternational audienceThe susceptibility of species to habitat changes depends on which ecological drivers shape individual fitness components. To date, only a few studies have quantified fitness components such as the Lifetime Reproductive Success across multiple generations in wild marine species. Because of a long-term sampling effort, such information is available for the population of wild orange clownfish, Amphiprion percula, from Kimbe Island (Papua New Guinea). Previous work on the wild orange clownfish near Kimbe Island suggests that there is little adaptive potential and that variation in LRS is mainly driven by a breeder’s habitat. Whether the host anemone species, geographic location, density or depth contributed to LRS remains however unknown because they were combined into a unique variable. We tested whether it is the ecology or the spatial distribution of clownfish that shaped the individual variation of a local fitness component, which would affect the population self-recruitment process and ultimately the maintenance of this wild population. Our spatially explicit analysis disentangled the role of these factors. We found that the host anemone species had an impact on wild clownfish LRS independently from their spatial distribution. The spatial distribution nevertheless had an impact on its own, as reflected by the spatial autocorrelation of LRS. Depth and density of anemones did not show a significant impact. Our findings imply that this clownfish population is susceptible to modifications of the spatial distribution and local assembly of anemone specie
Un examen de la procrastination dans une population multi-ethnique d'adolescents de Nouvelle-Calédonie
No full textInfographie résumant l'article scientifique suivant : Frayon S., Swami V., Wattelez G., Nedjar-Guerre A., Galy O. An examination of procrastination in a multi-ethnic population of adolescents from New Caledonia. BMC Psychol. 2023; 11:1. doi: 10.1186/s40359-022-01032-yInfography related to the following scientific paper: Frayon S., Swami V., Wattelez G., Nedjar-Guerre A., Galy O. An examination of procrastination in a multi-ethnic population of adolescents from New Caledonia. BMC Psychol. 2023; 11:1. doi: 10.1186/s40359-022-01032-yInfographie résumant l'article scientifique suivant : Frayon S., Swami V., Wattelez G., Nedjar-Guerre A., Galy O. An examination of procrastination in a multi-ethnic population of adolescents from New Caledonia. BMC Psychol. 2023; 11:1. doi: 10.1186/s40359-022-01032-
Les mécanismes de résilience des calédoniens pour préserver leur mode de vie face aux pénuries
No full textInternational audienceLes crises contemporaines, sanitaires, géopolitiques, environnementales ou sociales forcent les consommateurs à adapter leurs modes de vie. La Nouvelle-Calédonie n’échappe pas à ces crises et évolue, dans un système économique dépendant des importations (Ris, Trannoy et Wasmer, 2017). Le territoire connait régulièrement des pénuries et les populations adaptent leurs modes de vie afin d’en limiter les impacts sur leur mode de vie. Cette recherche tente d’appréhender les mécanismes de résilience mis en place par les consommateurs calédoniens face aux pénuries de biens de consommation courante, au-delà des habitudes déjà ancrées telles que stocker les produits de première nécessité (Guthrie, Fosso-Wamba, Arnaud, 2021) ou se reporter sur un produit plus cher (Khan et De Paoli, 2024). 1MéthodologieUne série d’entretiens en profondeur auprès de 20 consommateurs en face à face en Nouvelle-Calédonie a été réalisée autour des thèmes de la consommation quotidienne, des cognitions et émotions générées par les pénuries et des stratégies adaptatives mises en place. Des analyses lexicales et thématiques avec le logiciel NVivo ont été réalisées.2RésultatsL’analyse lexicale révèle trois champs lexicaux majeurs : (1) l’achat et la rupture : « acheter », « magasins », « besoin », ou « commerce », et « commander » « en rupture », « indisponible» ; (2) les stratégies adaptatives mises œuvre pour maintenir son mode de vie : « attendre », ou « faire du stock », « trouver » ; (3) le rapport au temp : le « temps » : « souvent », « jamais », « des années ». L’analyse thématique révèle trois thèmes principaux :(1)Le rôle du réseau social : les calédoniens s’appuient sur un tissu social fort. Ils font appel à leur réseau pour se tenir informés des livraisons via les réseaux sociaux. (2)L’option Consumer to Consumer (C to C) : Les individus trouvent des alternatives en dehors marché officiel et développent le troc, l'achat chez des personnes de leur entourage ou l'achat en direct chez des agriculteurs.(3)L’alternative « Consommerçant » (Lemaitre et de Barnier, 2015) : les individus sont tour à tour acheteur et/ou vendeur et deviennent « consommateur-producteur », en commercialisant leur production.3ConclusionNotre recherche montre que la population calédonienne est méfiante à l’égard du marché et des pratiques commerciales. Depuis longtemps exposées aux chocs et aux pénuries, elle combine plusieurs stratégies afin de développer sa résilience. La figure d’un « consom’acteur » calédonien émerge et rappelle celle du « prosumer » prophétisée par Toffler en 1980. L’auteur annonce la « démarchandisation » du monde et l’avènement d’un « consommateur-producteur » qui produit lui-même une grande partie des produits et services dont il a besoin, et les échanges le cas échéant. Le « consom’acteur » calédonien exprime sa résilience en alternant les différents rôles d’acheteur, de producteur et de vendeur, afin de reprendre le contrôle de sa consommation et de s’adapter au contexte changeant de l’économie. Ces résultats peuvent être rapprochés des travaux sur la consommation collaborative (Chameroy et al., 2024)
AI for SDGs – A technical and illustrated tour: Practical frameworks, real-world cases,and visual insights on harnessing AIto advance the UN 2030 Agenda
No full textInternational audiencePractical frameworks, real-world cases, and visual insights on harnessing AI to advance the UN 2030 Agenda How can Artificial Intelligence (AI) be effectively leveraged with the United Nations Sustainable Development Goals (SDGs)? Beyond the hype, AI has the potential to transform global challenges into opportunities—if applied responsibly and inclusively. Yet, connecting complex technical systems to urgent sustainability issues requires clarity, methodological rigor, and illustrative evidence. This book offers a structured and visually engaging exploration of how AI can support each of the 17 SDGs. By combining technical depth with accessible illustrations, it bridges the gap between advanced AI concepts and their practical applications in domains such as poverty estimation, climate action, health, and education. Readers will encounter real-world case studies, annotated diagrams, and examples that highlight both the promises and the limitations of AI for sustainability. Designed as both a reference and a guide, the book speaks to researchers, practitioners, policymakers, and students who want to understand not only the “what” and “why” but also the “how” of AI for sustainable development. By the end of this illustrated tour, readers will gain a clearer vision of where AI truly contributes, where caution is needed, and how innovation can be directed to serve the common good
SpaPool: Soft Partition Assignment Pooling for Graph Neural Networks
No full textInternational audienceThis paper introduces SpaPool, a novel pooling method that combines the strengths of both dense and sparse techniques for a graph neural network. SpaPool groups vertices into an adaptive number of clusters, leveraging the benefits of both dense and sparse approaches. It aims to maintain the structural integrity of the graph while reducing its size efficiently. Experimental results on several datasets demonstrate that SpaPool achieves competitive performance compared to existing pooling techniques and excels particularly on small-scale graphs. This makes SpaPool a promising method for applications requiring efficient and effective graph processing