HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Robust Confidence Intervals in Stereo Matching using Possibility Theory
No full textWe propose a method for estimating disparity confidence intervals in stereo matching problems. Confidence intervals provide complementary information to usual confidence measures. To the best of our knowledge, this is the first method creating disparity confidence intervals based on the cost volume. This method relies on possibility distributions to interpret the epistemic uncertainty of the cost volume. Our method has the benefit of having a white-box nature, differing in this respect from current state-of-the-art deep neural networks approaches. The accuracy and size of confidence intervals are validated using the Middlebury stereo datasets as well as a dataset of satellite images. This contribution is freely available on GitHub
Introduction au dossier « Santé mentale et milieux techniques »
No full textInternational audienceCe dossier thématique rassemble des contributions multiformes d’une partie des intervenants et intervenantes du séminaire PHITECO 2023 « Prendre soin de l’esprit : santé mentale et milieu technique », organisé par Elodie Gratreau, Florent Levillain et Vincenzo Raimondi (laboratoire Costech, équipe CRED), qui s’est tenu du 23 au 27 janvier 2023 à l’université de technologie de Compiègne. Les contributions explorent tour à tour des trajectoires historiques (Dupont ; Guillemain ; Dubus) et des situations contemporaines (Alombert ; Gagné-Julien), ainsi qu’elles interrogent l’élaboration des savoirs (Moutaud ; Gratreau) et les pratiques de conception (Gaboriau ; Guénand)
The use of Advanced Numerical Models in The Optimization of Embedding Position of d33 Piezoelectric Transducers
No full textInternational audienceIn this study, the best embedding position of piezoelectric sensors through the thickness of composite materials was investigated for structural health monitoring purposes via Lamb wave propagation. The work encompasses both numerical and experimental approaches towards achieving the desired optimal position. The numerical work utilizes a kinematically refined 2D FE plate model based on the Carrera Unified Formulation, offering a compelling computational cost compared to the full 3D FE models. The numerical results cover a wide scope on the different embedding possibilities whereas the experimental results serve as a qualitative verification
A Global-Local Modelling Approach for Wave Propagation, SHM and Damage Detection in Reinforced Panels Using Integrated Piezoelectric Sensors
No full textInternational audienceThis paper presents forward a novel method for the damage detection in smart structures. The use of node-dependent kinematic models, based on the Carrera Unified Formulation, has led to an innovative global-local approach, with the ability to couple finite elements with different kinematics, and thus different levels of accuracy. With such approach it is possible to model a composite structure with embedded piezoelectric transducer, where the integrated patch or region is modelled using high-fidelity models, as the complex phenomena is expected to happen locally around that area. Thus, the use of layer-wise model is necessary on the level where higher accuracy is needed, as in the embedded area, whereas it remains possible to implement the cost efficient homogenized equivalent single layer models in other areas, reducing the computational cost. This approach has been used to study the lamb-wave propagation in reinforced composite panels, where a damage has been introduced. The detection of such damage through pitch and catch set-up has been studied. The reflected signal due to damage has also been investigated. The results show that this approach may provide accurate damage detection localization and accurate wave propagation despite the presence of localized models with sensors/ actuators
Two experiments and a new model for the digital processing of Elisa Chimenti's corpus (Naples 1883-Tangier 1969)
No full textInternational audienceThis paper presents two digitisation workflows and a new model proposition for the digital processing of unpublished texts by Elisa Chimenti (1883-1969). The author's corpus is extremely varied and interesting both from the point of view of academic research and for the wider public. The experiments therefore move in the area of open access digital scholarly edition. The aim is to compare the different text processing chains, in order to reflect on the pros and cons of each. Further, the new proposition aims to build a model for the digitisation of this corpus, also focusing on ethical and ecological issues that contemporary society confronts us with. The purpose is to contribute to the ongoing project of organising, preserving, and digitising the heritage of Elisa Chimenti, an extraordinary as well as forgotten author
On P-refinement in Topology Optimization
No full textInternational audienceStructural topology optimization is carried out by optimizing relative density; i.e., design variable in each element is refined by refining the FE mesh size, where the number of density elements is scaled up beside the increase in interpolation order of finite elements. Higher-order elements can be used to obtain more realistic solutions compared to solutions with lower-order elements which includes the stability of numerical solution. In this technical script, dominance of p-refinement in topology optimization by solid isotropic material with penalization includes compliance values and computational time for various topology optimization problems
Où sont passées mes clefs ? : Plausibilité et cohérence des scénarios guidant la recherche d’objets égarés
No full textInternational audienceHow we proceed when searching for a lost object is rarely haphazard : we rely on search strategies, considering different hypotheses about where the object might be, and gradually eliminating them until we locate it. To study the reasoning processes at work in this type of activity, we subjected participants to a lost object searching task in an ecological context and analyzed the heuristics spontaneously used to organize the search activity. The results show that the strategies are constrained by informational coherence and plausibility. The participants, to explain the disappearance of the object and organize their research activity, favor scenarios which appear plausible to them regarding the elements of information available, especially memorized facts, and which are compatible, more generally, with their knowledge of the general principles governing how objects change position in space.La manière dont nous procédons lorsque nous cherchons un objet égaré est rarement anarchique : nous mettons en œuvre des stratégies de recherche, envisageant différentes hypothèses sur l’endroit où pourrait se trouver l’objet, et les éliminant progressivement jusqu’à localiser celui-ci. Pour étudier les processus de raisonnement à l’œuvre dans ce genre d’activités, nous avons soumis des participants à une tâche de recherche d’objets en contexte écologique et avons analysé les heuristiques spontanément mises en œuvre pour organiser l’activité de recherche. Les résultats montrent que les stratégies employées sont contraintes par des exigences de cohérence informationnelle et de plausibilité. Les participants, pour s’expliquer la disparition de l’objet et organiser leur activité de recherche, favorisent des scénarios qui leur apparaissent plausibles au regard des éléments d’information disponibles, en particulier les faits mémorisés, mais qui sont compatibles, plus généralement, avec leur connaissance des principes généraux régissant le déplacement des corps
Uniform in number of neighbors consistency and weak convergence of NN empirical conditional processes and NN conditional -processes involving functional mixing data
No full textInternational audience-statistics represent a fundamental class of statistics arising from modeling quantities of interest defined by multi-subject responses. -statistics generalize the empirical mean of a random variable to sums over every -tuple of distinct observations of . Stute [182] introduced a class of so-called conditional -statistics, which may be viewed as a generalization of the Nadaraya-Watson estimates of a regression function. Stute proved their strong pointwise consistency to: r^{(m)}(\varphi, \mathbf{t}): = \mathbb{E}[\varphi(Y_{1}, \ldots, Y_{m})|(X_{1}, \ldots, X_{m}) = \mathbf{t}], \; \mbox{for}\; \mathbf{ t}\in \mathcal{X}^{m}. In this paper, we are mainly interested in the study of the NN conditional -processes in a functional mixing data framework. More precisely, we investigate the weak convergence of the conditional empirical process indexed by a suitable class of functions and of the NN conditional -processes when the explicative variable is functional. We treat the uniform central limit theorem in both cases when the class of functions is bounded or unbounded satisfying some moment conditions. The second main contribution of this study is the establishment of a sharp almost complete Uniform consistency in the Number of Neighbors of the constructed estimator. Such a result allows the number of neighbors to vary within a complete range for which the estimator is consistent. Consequently, it represents an interesting guideline in practice to select the optimal bandwidth in nonparametric functional data analysis. These results are proved under some standard structural conditions on the Vapnik-Chervonenkis classes of functions and some mild conditions on the model. The theoretical results established in this paper are (or will be) key tools for further functional data analysis developments. Potential applications include the set indexed conditional U-statistics, Kendall rank correlation coefficient, the discrimination problems and the time series prediction from a continuous set of past values.</p
Faire dire, faire taire : la modération (algorithmique) des réseaux sociaux numériques au prisme des sexualités
No full textThis thesis explores the challenges of content moderation on digital social media. Itstarts from examining the paradox of using AI systems and algorithms as aids in the mod-eration process, which can inadvertently silence certain marginalized voices and identitiesonline. Through a reinterpretation of the concepts of dispositive de sexualité and disposi-tif d’écriture, the thesis postulates that moderation shapes, silences, and reshapes sexualand gender identities online. The thesis also gives significant attention to digital methodsand « savoirs d’expérience » in the study of digital devices : how to collect and analyzedeleted content on social social media (such as Twitter/X and TikTok in particular)? Fi-nally, I focus on the algorithmic circumvention tactics employed by online gay communities to distribute sexually explicit content on social media that claim to prohibit it. Drawingon Anglo-Saxon references in porn, gender, and post-colonial studies, the thesis also con-tributes to the circulation of a number of English-language works on algorithmic bias andonline discrimination and helps define platform affordances in the study of digital devices.The thesis adheres to the following outline : a first part composed of two chapters presentswhat constitutes content moderation (between human labor, machines, and texts) and whatdistinguishes it from platform regulation. In this first part, I also explore a little-knownhistory of the internet through the prism of gender and sexuality. The second part invites tothink about content moderation as the custodian of the dispositif de sexualités and disposi-tif d’écriture online. Through reinterpretations of Michel Foucault’s concept of dispositif desexualité, I propose to consider how moderation shapes, silences, and reshapes sexualitiesonline. This part also presents all the methodological apparatus set up with computer en-gineers at CERES (Sorbonne University) to collect and analyze content deleted by contentmoderation. I also present several methodological leads for studying the role of algorithmsin the moderation process. To overcome the opacity of these digital devices, I give signifi-cant importance to « savoirs d’expérience » in the construction of research hypotheses. Inparticular, I return to the hypothesis of an invisibility of sexual and gendered communitiesonline, due to the keywords that would be used to designate these identities - and that al-gorithms would mistakenly classify as abusive or exclusively pornographic content. Finally,this part includes a self-reflective exploration of the use of digital methods in humanitiesscholarship. The third part presents the results of the study, focusing on the specific casesof Twitter/X and TikTok. The modalities of expression of sexual and gendered communitiesonline, the automated deletion of content by keyword on Twitter, and the circumventiontactics implemented by gay internet users and sex workers to post sexually explicit contenton TikTok are the main research tracks studied in this part. In short, this part invites us toconsider the modalities of adjustment of the closet of sexual and gendered communities tobecome (in)visible online.La thèse porte sur les enjeux de modération de contenus dans les réseaux sociaux numé-riques. Le point de départ de la thèse est d’explorer le paradoxe qu’il peut exister dans lerecours aux systèmes d’IA et aux algorithmes comme des aides au processus de modérationet qui invisibilisent certains discours et identités minorisées en ligne. A travers une relecturedes notions de dispositifs de sexualités et d’écriture, la thèse postule que la modération faitdire, fait taire et fait dire autrement les identités sexuelles et de genre en ligne. La thèse faitégalement la part belle aux méthodes numériques et aux savoirs d’expérience dans l’étudedes dispositifs numériques : Comment collecter et analyser des contenus supprimés dans lesRSN (comme Twitter/X et TikTok en particulier)? A partir de références anglosaxonnes enporn studies et en études décoloniales, la thèse participe aussi à la circulation d’un certainnombre de travaux anglophones sur les biais et discriminations algorithmiques en ligne, etcontribue à la définition des affordances de plateforme dans l’étude des dispositifs numé-riques. La thèse suit donc le plan suivant : une première partie composée de deux chapitresprésente ce qui constitue la modération des réseaux sociaux (entre travail humain, machineset textes) et ce qui la distingue de la régulation des plateformes. Dans cette première partie,j’explore également une histoire méconnue d’internet au prisme du genre et des sexualités.La seconde partie invite à penser la modération de contenus comme la garante du disposi-tif de sexualités et d’écriture des réseaux sociaux numériques. A travers des relectures dela notion de dispositif de sexualité de Michel Foucault, je propose de considérer commentla modération fait dire, fait taire et fait dire autrement les sexualités en ligne. Cette par-tie présente également tout l’appareillage méthodologique mis en place avec des ingénieursen informatique du CERES (Sorbonne Université) pour collecter et analyser des contenussupprimés par la modération de contenus. Je présente aussi un certain nombre de pistes mé-thodologiques pour étudier la part de recours aux algorithmes dans le travail de modération. Pour pallier l’opacité de ces dispositifs numériques, j’accorde une place importante aux sa-voirs d’expérience dans la construction d’hypothèses de recherche. En particulier, je revienssur l’hypothèse d’une invisibilisation des communautés sexuelles et de genre en ligne, enraison des mots-clés qui seraient utilisés pour désigner ces identités – et, que les algorithmesassimileraient, à tort, à des contenus injurieux ou exclusivement pornographiques. Enfin,cette partie comporte un retour réflexif sur la place des méthodes numériques dans l’envi-ronnement de la recherche en SHS. La troisième partie présente les résultats de l’étude en sefocalisant sur les terrains de Twitter/X et TikTok en particulier. Les modalités d’expressiondes communautés sexuelles et de genre en ligne, la suppression automatisée de contenus parmot-clé sur Twitter et les tactiques de contournement mises en place par les internautesgay et les travailleurs du sexe pour poster des contenus à caractère sexuel sur TikTok sontles principales pistes de travail étudiées dans cette partie. En somme, cette partie invite àconsidérer les modalités d’ajustement du placard des communautés sexuelles et de genrepour se rendre (in)visible en ligne
