Portail des publications scientifiques IMT Mines Alès
Not a member yet
5198 research outputs found
Sort by
FineFreq: A Multilingual Character Frequency Dataset from Web-Scale Text
We present FineFreq, a large-scale multilingual character frequency dataset derived from the FineWeb and FineWeb2 corpora, covering over 1900 languages and spanning 2013-2025. The dataset contains frequency counts for 96 trillion characters processed from 57 TB of compressed text. For each language, FineFreq provides per-character statistics with aggregate and year-level frequencies, allowing fine-grained temporal analysis. The dataset preserves naturally occurring multilingual features such as cross-script borrowings, emoji, and acronyms without applying artificial filtering. Each character entry includes Unicode metadata (category, script, block), enabling domain-specific or other downstream filtering and analysis. The full dataset is released in both CSV and Parquet formats, with associated metadata, available on GitHub and HuggingFace. https://github.com/Bin-2/FineFre
Effect of glass powder on the physico-mechanical and microstructural properties and durability of calcined clay-based geopolymer binder
International audienceIn sub-Saharan Africa, and more specifically in Burkina Faso, geopolymer binders are often synthesized from calcined clays. This method is preferred because industrial by-products, such as fly ash, blast furnace slag, and red mud, are scarcely available, unlike those in more industrialized countries. However, the clay calcination process is energy-intensive. To address this issue, the use of noncalcined precursor additives has been widely explored. Glass powder is a promising alternative. The amorphous silica content can promote geopolymerization by supplementing the silica content of the calcined clay. The objective of this study was to evaluate the physico-mechanical behavior and durability of a geopolymer binder in which calcined clay (metakaolin, M) was replaced with glass powder (P) at various mass ratios (from 0% to 25%) and activated with a 12 mol/L NaOH solution. The formulated binders were cured for 14 d, with the first 7 d at ambient temperature (30 ± 5 °C) and the remaining 7 d of thermal curing at 60 ± 5 °C in an oven. The characterization results showed a significant improvement in the engineering properties and durability of the geopolymer binders, depending on the substitution ratio of calcined clay with glass powder. The relative improvements in various properties at the optimal glass powder content (20%, specifically 80M20P) compared to the matrix without glass powder (0%, specifically 100M0P) included a 6% increase in bulk density, 8.9% decrease in water-accessible porosity, 533% and 166% increase in flexural and compressive strength, respectively, and 38% decrease in mass loss after 28 days of acid attack. This suggests that substituting calcined clay and adding up to 20% can achieve even better engineering and durability behaviors in geopolymer binders, thereby adding value to glass waste
Bridging MBSE and MBSA through an Interoperability Framework
International audienceModel-based systems engineering (MBSE) involves the implementation and intensive use of models to perform systems engineering. Similarly, model-based safety analysis or assessment (MBSA) is defined as an effort to carry out dependability engineering,as document-centric methodologies do not provide an efficient way of dealing with complex system development challenges. Although model-based engineering approaches such as MBSE and MBSA have been developed, the exchange of information betweenthe two domains is primarily based on documents. This document-centric interaction raises numerous issues regarding the consistency and efficiency of the overall modeling. A model-based interoperability framework is necessary to overcome this limitation.In this paper, the authors propose a unified engineering framework (UEF) as an initiative to bridge the gap between system models,such as MBSE and MBSA models. An application use case in the nuclear sector is also introduced to show the practical applicationand benefits of developing this framework
Optimisation d'un procédé d'extraction de fibres végétales et synthétiques
National audienc
Exercices et préparation des équipes à la réponse opérationnelle en cas de situation exceptionnelle. Intérêts et limites
National audienc
Elicitation des hyper-paramètres des classifieurs prudents : premières propositions
National audienceIn the presence of uncertainty, cautious prediction methodsare essential to avoid serious errors when using machinelearning in sensitive domains, e.g., medicine, autonomousvehicles, etc. These methods involve the useof utility or loss functions in the decision and/or evaluationstages. The utility function, which is inherentlysubjective, extends the classical and inherently objectiveaccuracy metric to comparisons of subsets. These functionsare parameterized to express the user’s dispositionfor imprecision. This article presents preliminary workaddressing the underexplored problem of eliciting theseparameters from the user. We show how the user is guidedthrough successive comparisons toward the optimalvalue.En pr´esence d’incertitude, les m´ethodes de pr´edictionprudente sont essentielles pour ´eviter des erreurs graveslors de l’utilisation de l’apprentissage automatique dansdes domaines sensibles, e.g. m´edecine, voiture autonome,etc. Ces m´ethodes impliquent l’utilisation de fonctiond’utilit´e ou de perte dans l’´etape de d´ecision et/ou l’´etaped’´evaluation. La fonction d’utilit´e en question, par essencesubjective, ´etend la m´etrique classique de justesse,par essence objective, aux comparaisons de sousensembles.Ces fonctions sont param´etr´ees pour exprimerla disposition de l’utilisateur `a l’impr´ecision. Cetarticle propose un travail pr´eliminaire qui s’int´eresse auprobl`eme sous explor´e de l’´elicitation de ces param`etresavec l’utilisateur. Nous montrons comment l’utilisateurest guid´e par des comparaisons successives jusqu’`a la valeuroptimale
Évaluation de prédictions prudentes dans le cas d'observations partielles
National audienceIn different situations, one will have to evaluate setvaluedpredictions while having set-valued/partial observations.This is the case, for instance, when experts provideweak supervision and when predictions are providedby robust classifiers. However, we still miss some principledapproach to evaluate such predictions and observations.This is the goal of this paper, where we proposeguidelines about such evaluations functions. We will setour study within the framework of belief functions, thatseem quite adapted to the considered problem.Dans différentes situations, il est nécessaire d’évaluer des prédictions prudentes sous forme d’ensemble de classes ou d’intervalles tout en disposant d’observations également prudentes ou partielles. C’est le cas, par exemple, lorsque des experts fournissent une supervision faible, ou lorsque les prédictions proviennent de classifieurs robustes. Cependant, il manque encore une approche rigoureuse pour évaluer de telles prédictions en présence de telles observations. C’est l’objectif de cet article, dans lequel nous proposons des lignes directrices pour la définition des fonctions d’évaluation. Nous inscrivons notre étude dans le cadre des fonctions de croyance, qui semblent particulièrement bien adaptées au problème considéré
Strain-dependent relaxation time of viscoelastic soft tissues
International audiencePerineal tears occur in about 80 % of vaginal deliveries. During childbirth, the descent of the infant causes repeated stretching of the perineal body. The tissues of the perineum undergo significant elongation, with the stretch maintained until the next pushing effort, which induces an even greater level of stretch—reaching up to 400% elongation. This sequence of stretching and relaxation is repeated several times, potentially causing tissue failure and leading to perineal lacerations, which may result in pelvic floor disorders.Conservative techniques exist to alleviate perineal stretch, such as maneuvers or episiotomy. The latter involves a controlled incision in the perineal body. However, these techniques can sometimes cause more harm than the natural occurrence of tearing, particularly if the risk of deep lacerations during delivery is overestimated. Therefore, it is crucial to develop a non-destructive method for assessing perineal damage to support physicians in making informed decisions during childbirth.To address this, the mechanical response of the perineal body to repeated elongation-relaxation loads was investigated via cyclic pure-shear testing conducted ex vivo on porcine perinea. Analysis of the relaxation phase using Zener model identification revealed that the relaxation times continuously decreased with increased stretching until a discontinuity occurred. This discontinuity appeared to precede visible structural damage (tears) in the sample.These findings suggest that the evolution of the viscoelastic response of the perineal tissue can be measured non-invasively, potentially using relaxed indentation or suction testing. The results are promising for the future development of real-time assessments of the mechanical state of the perineum during childbirth
From the Surface to the Underground : Towards future interstitial public spaces. Casestudy of three Grand Paris Express stations
International audienceThe Grand Paris Express underground transport project is reconfiguringspaces in the Paris region. With 68 new stations designed by architects,these infrastructures are no longer mere transit points, but spatial andsocial interfaces between surface and the underground. This paper offers amulti-scalar reading of these spaces, both spatially (verticality and in-depth)and temporally (from conception to use), for three stations: Villejuif-GustaveRoussy, Clichy-Montfermeil and Noisy-Champs. The documents producedby architects (plans, sections, perspectives, models) are not only tools ofrepresentation but are also ideal projections of a desired future (Choay,1965). As a moment of imagination, these spatial productions articulate time:past, present and future.These documents will be compared with a series of sensitive sketchescarried out in and around the stations, to cross-reference the projectedintention with situated perception. Documenting the actual uses, functionalconstraints or social conflicts that run through these places, allows theformulation of a critical look at the projects‘ visions. Drawing on the conceptof the critical zone (Latour, Weibel, 2020)1, this work examines these stationsas interstitial public places between surface and subsoil, where tensionsbetween technical infrastructures, social life and environment materialize.Through a visual and comparative analysis, the aim of this study is to shiftthe way we look at these infrastructures, considering them as stratifiedenvironment where forms of urban, material and symbolic cohabitation arenegotiated and renewed daily