HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Credal ensembling in multi-class classification
No full textInternational audienceIn this paper, we present a formal framework to (1) aggregate probabilistic ensemble members into either a representative classifier or a credal classifier, and (2) perform various decision tasks based on this uncertainty quantification. We first elaborate on the aggregation problem under a class of distances between distributions. We then propose generic methods to robustify uncertainty quantification and decisions, based on the obtained ensemble and representative probability. To facilitate the scalability of the proposed framework, for all the problems and applications covered, we elaborate on their computational complexities from the theoretical aspects and leverage theoretical results to derive efficient algorithmic solutions. Finally, relevant sets of experiments are conducted to assess the usefulness of the proposed framework in uncertainty sampling, classification with a reject option, and set-valued prediction-making
Rates in the complete convergence of weighted bootstrap means
No full textInternational audienceIn the present article, we consider the almost sure conditional complete convergence of weighted bootstrap means analogously to that obtained by Csörgő (Citation2003) for the multinomial bootstraps. More precisely, in a similar fashion to Gut and Spătaru’s theorems for ordinary means, precise asymptotic behavior is extended to weighted bootstrap means provided that the bootstrap sample sizes are the integer part of any fixed power of the original sample sizes
Robot autonome pour l'inspection acoustique de sites industriels : théorie et expériences en temps réel
No full textDrones et taxi volants en milieu urbain; GAHA - Aéro et Hydro-Acoustique: GABE - Acoustique du Bâtiment et de l'EnvironnementNational audienceLa localisation précise des sources sonores est cruciale pour des diagnostics acoustiques efficaces dans des environnements industriels réels. Cette étude comble un vide expérimental dans le domaine de l'audition des robots en introduisant une expérience en temps réel qui démontre une planification de trajectoire efficace pour identifier de manière autonome de multiples sources sonores dans un environnement inconnu. Nous présentons un cadre expérimental complet pour valider des algorithmes de contrôle avancés basés sur des données acoustiques pour guider le mouvement du robot. En outre, une méthode de filtrage spatial est introduite pour gérer les sources sonores précédemment détectées. L'approche est d'abord évaluée par simulation, démontrant son potentiel théorique, et est ensuite testée dans une expérience en temps réel où un véhicule volant autonome équipé d'un réseau de 32 microphones MEMS localise avec succès deux sources sonores inconnues dans un environnement intérieur
Guaranteed prediction sets for functional surrogate models
No full textInternational audienceWe propose a method for obtaining statistically guaranteed prediction sets for functional machine learning methods: surrogate models which map between function spaces, motivated by the need to build reliable PDE emulators. The method constructs nested prediction sets on a low-dimensional representation (an SVD) of the surrogate model's error, and then maps these sets to the prediction space using set-propagation techniques. This results in prediction sets for functional surrogate models with conformal prediction coverage guarantees. We use zonotopes as basis of the set construction, which allow an exact linear propagation and are closed under Cartesian products, making them well-suited to this high-dimensional problem. The method is model agnostic and can thus be applied to complex Sci-ML models, including Neural Operators, but also in simpler settings. We also introduce a technique to capture the truncation error of the SVD, preserving the guarantees of the method
Discrete Minimax Probabilistic Classifier Chains for Multi-label Classification Under Label Imbalance
No full textInternational audienceIn multi-label classification (MLC), each instance can be assigned tonone, one or multiple labels from a predefined label set, and the task is to predictthe relevant subset of labels for each new instance. A common challenge in MLCis class imbalance, which often leads to biased classifiers that underestimate rarelabels. In this paper, we propose a framework that combines probabilistic classifier chains (PCC) with a minimax learning strategy based on the Discrete minimaxclassifier. PCC allows us to model label dependencies and optimize various lossfunctions, including the subset 0/1 loss, Hamming loss, and F1-measure, while theminimax learning strategy mitigates the class imbalance problem by minimizingand balancing the class conditional risks. We conduct experiments on ten benchmark datasets using multiple models of different learning strategies. Our analysisfocuses on the ability of models to optimize a loss function while also reducingthe false positive and false negative rates. To this end, we use two complementarymetrics designed to measure imbalance in class-conditional accuracies
Semantic History of Medieval Music: A Wikidata Framework for Computational Historiography: A Wikidata-Grounded Framework for Computational Historiography
No full textCe rapport présente le projet Semantic History of Medieval Music, un cadre conceptuel et computationnel pour une historiographie sémantique et ouverte de la musique médiévale (500–1500). Basé sur Wikidata et les principes du Linked Open Data, il combine graphes multilingues, embeddings et RAG (Retrieval-Augmented Generation) pour explorer les continuités entre concepts musicaux, manuscrits et systèmes modaux à travers les cultures
Cost Optimization in a GI/M/2/N Queue with Heterogeneous Servers, Working Vacations, and Impatient Customers via the Bat Algorithm
No full textInternational audienceThis paper analyzes a finite-capacity GI/M/2/N queue with two heterogeneous servers operating under a multiple working-vacation policy, Bernoulli feedback, and customer impatience. Using the supplementary-variable technique in tandem with a tailored recursive scheme, we derive the stationary distributions of the system size as observed at pre-arrival instants and at arbitrary epochs. From these, we obtain explicit expressions for key performance metrics, including blocking probability, average reneging rate, mean queue length, mean sojourn time, throughput, and server utilizations. We then embed these metrics in an economic cost function and determine service-rate settings that minimize the total expected cost via the Bat Algorithm. Numerical experiments implemented in R validate the analysis and quantify the managerial impact of the vacation, feedback, and impatience parameters through sensitivity studies. The framework accommodates general renewal arrivals (GI), thereby extending classical (M/M/2/N) results to more realistic input processes while preserving computational tractability. Beyond methodological interest, the results yield actionable design guidance: (i) they separate Palm and time-stationary viewpoints cleanly under non-Poisson input, (ii) they retain heterogeneity throughout all formulas, and (iii) they provide a cost–optimization pipeline that can be deployed with routine numerical effort. Methodologically, we (i) characterize the generator of the augmented piecewise–deterministic Markov process and prove the existence/uniqueness of the stationary law on the finite state space, (ii) derive an explicit Palm–time conversion formula valid for non-Poisson input, (iii) show that the boundary-value recursion for the Laplace–Stieltjes transforms runs in linear time O(N) and is numerically stable, and (iv) provide influence-function (IPA) sensitivities of performance metrics with respect to (μ1,μ2,ν,α,ϕ,β)
Large-scale separation of gram-level tire and road wear particles from road dust
No full textInternational audienceSeparation approaches have been developed to isolate tire and road wear particles (TRWPs) from environmental media, demonstrating their potential in providing representative test materials for TRWPs studies. However, the achievable mass of TRWPs and the broader applicability of the method have yet to be verified. This study aims to determine whether separation methods can acquire gram-scale TRWPs from various road dust samples. A large-scale dust separation protocol incorporating two-stage density separation and software-assisted TRWPs selection was adapted from existing methods. Twenty-five road dust samples from various road types and regions in a megacity were collected and processed using this protocol. The selected particles were identified as TRWPs using SEM-EDX, FTIR, and Py-GC–MS. Gram-scale yields (>1.0 g TRWPs per kg dust) were achieved at 44 % of sampling sites. Additionally, we proposed three recommended road dust sampling locations and estimated that collecting 1.3–2.3 kg of dust from these sites would yield one gram of TRWPs. This is the first demonstration that gram-level TRWPs can be reliably obtained from field dust, enabling environmental fate studies. Future research should focus on the standardization of environmental TRWPs as test materials
Les plateformes numériques dans le secteur agricole : vers une durabilité par le partage des connaissances
No full textAs digital transformation continues to reshape the modern economy, digital platforms are emerging as dynamic with transformative potential, particularly within the agricultural sector. At a time when agriculture is being called to reinvent its production methods as a response to growing sustainability challenges, this research explores the role of digital platforms in facilitating a transition towards sustainable agriculture. The main objective of this research is to explore how digital platforms can structure and enhance knowledge networks, bridge geographic and institutional divides among agricultural stakeholders, and foster the adoption of sustainable practices. Drawing on a multiple qualitative approach, grounded in the conceptual framework of “digital sustainability” and actor-network theory, this research uncovers both the opportunities these platforms present and the risks and limitations they entail. Three main themes structure our analysis: the conditions required for the development of digital platforms, their potential contribution to knowledge-sharing dynamics, and the challenges associated with their integration into diverse agricultural contexts. This research lies at the crossroads of current discussions on the convergence between digitalization and sustainability. It offers both theoretical insights and practical guidance for designing platforms that are rooted in the realities of the agricultural landscape and aligned with global sustainable development goals.Face aux mutations numériques qui redéfinissent les contours de l’économie contemporaine, les plateformes numériques émergent comme des écosystèmes dynamiques au potentiel transformateur, notamment dans le secteur agricole. Dans un contexte où l’agriculture doit réinventer ses modes de production pour relever les défis de durabilité, cette thèse interroge le rôle de ces dispositifs numériques comme leviers pour accompagner la transition vers une agriculture durable. L’objectif principal de cette recherche est d’explorer comment ces plateformes peuvent structurer et valoriser les flux de connaissances, réduire les distances géographiques et institutionnelles entre les acteurs agricoles, et soutenir des pratiques agricoles durables. En mobilisant une approche qualitative multiple, le cadre conceptuel de la « durabilité numérique » et la théorie de l’acteur-réseau, cette thèse met en lumière le potentiel des plateformes numériques pour catalyser la transition agricole, tout en révélant leurs risques et limites. Trois axes d’interrogation structurent la réflexion de cette thèse : les conditions nécessaires au déploiement des plateformes numériques, leur contribution aux dynamiques de partage de connaissances, et les défis liés à leur intégration dans des contextes agricoles diversifiés. Inscrite au cœur des débats actuels sur la convergence entre numérisation et durabilité, cette recherche propose des perspectives théoriques et pratiques pour développer des plateformes ancrées dans les réalités agricoles et alignées sur les objectifs du développement durable
k-Nearest Neighbour Estimation of the Conditional Set-Indexed Empirical Process for Functional Data: Asymptotic Properties
No full textInternational audienceThe main aim of this paper is to improve the existing limit theorems for set-indexed conditional empirical processes involving functional strong mixing random variables. To achieve this, we propose using the k-nearest neighbor approach to estimate the regression function, as opposed to the traditional kernel method. For the first time, we establish the weak consistency, asymptotic normality, and density of the proposed estimator. Our results are derived under certain assumptions about the richness of the index class C, specifically in terms of metric entropy with bracketing. This work builds upon our previous papers, which focused on the technical performance of empirical process methodologies, and further refines the prior estimator. We highlight that the k-nearest neighbor method outperforms the classical approach due to several advantages