Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Localized integration of iron-based nanoparticle micromagnets on planar inductors for RF applications
No full textInternational audienceHigh permeability magnetic cores are often used in electronic components to concentrate electromagnetic field lines in order to enhance their radiofrequency properties or to reduce the device size while maintaining their performances. For that purpose, efficient magnetic materials are required, with major magnetic properties, dielectric behavior and soft techniques of integration. Here we present a new bottom-up approach for integrating localized magnetic dielectric materials onto planar inductors. Magnetic nanoparticles are first synthesized by a liquid-phase chemical approach and then assembled into submillimeter magnets onto inductors and coplanar waveguides by a magnetophoresis-directed process, performed at atmospheric pressure and room temperature. This approach allows a conformal deposition while limiting material losses. Permittivity and permeability of Fe, FeC and FeCo particles based-materials are extracted up to 30 GHz. Optimizing the magnetic properties of the magnets through the size and chemical composition of the nanoparticles allow increasing of 40 to 60% the nominal inductance values of 2 and 13 nH up to 3 GHz. These results open new perspectives for the design of radio-frequency electronic micro-components
André Villiers, 1951-1955 : le théâtre universitaire américain, un modèle pour la création d’études théâtrales à la Sorbonne ?
No full textInternational audienc
Quand Lincoln et Booth parlent verlan : quelques considérations relatives à la traduction de "Topdog/Underdog", de Suzan-Lori Parks, suivies de la performance live du fameux "3-Card Monte scam"
No full textInternational audienc
Turbo-Muon: Accelerating Orthogonality-Based Optimization with Pre-Conditioning
No full textOrthogonality-based optimizers, such as Muon, have recently shown strong performance across large-scale training and community-driven efficiency challenges. However, these methods rely on a costly gradient orthogonalization step. Even efficient iterative approximations such as Newton-Schulz remain expensive, typically requiring dozens of matrix multiplications to converge. We introduce a preconditioning procedure that accelerates Newton-Schulz convergence and reduces its computational cost. We evaluate its impact and show that the overhead of our preconditioning can be made negligible. Furthermore, the faster convergence it enables allows us to remove one iteration out of the usual five without degrading approximation quality. Our publicly available implementation achieves up to a 2.8× speedup in the Newton-Schulz approximation. We also show that this has a direct impact on end-to-end training runtime with 5-10% improvement in realistic training scenarios across two efficiency-focused tasks. On challenging language or vision tasks, we validate that our method maintains equal or superior model performance while improving runtime. Crucially, these improvements require no hyperparameter tuning and can be adopted as a simple drop-in replacement. Our code is publicly available on github
Bony labyrinth morphology reveals deep population history and isolation in prehistoric northeastern Africa (44–3 ka)
No full textInternational audienc
A tight relationship between BOLD fMRI activation/deactivation and increase/decrease in single neuron responses in human association cortex
No full textInternational audienceThe relationship between Blood-Oxygen-Level-Dependent (BOLD) responses in functional magnetic resonance imaging (fMRI) and increases or decreases in neural firing rate across human brain regions, especially the association cortex, remains largely unknown. Here, we contrast direct measures of neuronal activity in two adjacent brain regions of the fusiform gyrus (FG) associated with fMRI increases (lateral FG portion) or decreases (medial FG portion) of the same category-selective neural activity. In both individual brains tested across multiple recording sessions, a frequency-tagging stimulation objectively identified a substantial proportion (about 70%) of face-selective neurons. While single-units recorded in the lateral FG showed a selective increase to faces, neurons localized in the medial FG decreased spiking activity selectively to faces. Beyond a relative reduction to faces compared to nonface objects, about a third of single neurons found in the medial FG showed genuine suppression of baseline spiking activity upon presentation of a face. These observations clarify the nature of face-selective neural activity in the human brain, which can be expressed both as increases and active suppressions of spiking activity, and, more generally, shed light on the physiological basis of the fMRI signal
Mesurer l’impact du tourisme équitable et solidaire
No full textFair and solidarity-based tourism aims to transform relationships between travellers and host communities by promoting participatory, redistributive practices rooted in local development. Yet its actual economic impacts remain difficult to assess, limiting its institutional recognition. This thesis explores the conditions required to implement a dedicated observatory capable of making the model’s positive economic effects more visible. Based on a professional placement within the association Rencontres au Bout du Monde, the research proposes a comprehensive methodological framework combining an adapted analysis grid, simple tools, and shared data governance. It concludes with a proposal for a lightweight, ethical and transferable observatory model designed for solidarity tourism actors.Le tourisme équitable et solidaire s’inscrit dans une démarche de transformation des rapports entre voyageurs et communautés d’accueil, en promouvant des pratiques participatives, redistributives et ancrées dans le développement local. Pourtant, ses effets économiques réels restent difficiles à mesurer, ce qui freine sa reconnaissance institutionnelle. Ce mémoire interroge les conditions nécessaires à la mise en place d’un observatoire spécifique, capable de rendre visibles les retombées économiques positives du modèle. À partir d’une immersion en alternance au sein de l’association Rencontres au Bout du Monde, il propose une démarche méthodologique complète, fondée sur une grille d’analyse adaptée, des outils simples, et une gouvernance partagée de la donnée. Le mémoire débouche sur une modélisation d’observatoire léger, reproductible et éthique, à destination des acteurs du tourisme solidaire
Set-Valued Koopman Theory for Control Systems
No full textInternational audienceIn this paper, we introduce a new notion of Koopman operator which faithfully encodes the dynamics of controlled systems by leveraging the grammar of set-valued analysis. In this context, we propose meaningful generalisations of the Liouville and Perron-Frobenius operators, and show that they respectively coincide with proper set-valued analogues of the infinitesimal generator and dual operator of the Koopman semigroup. We also give meaning to the spectra of these set-valued maps and prove an adapted version of the classical spectral mapping theorem relating the eigenvalues of a semigroup and those of its generator. In essence, these results provide theoretical justifications for existing approaches in the Koopman communities which consist in studying control systems by bundling together the Liouville operators associated with different input parameters
HIGH-THROUGHPUT MECHANOBIOLOGICAL CELL DISCRIMINATION USING AUTOMATED AFM AND MACHINE LEARNING
No full textInternational audienceMechanobiological measurements offer a promising avenue for distinguishing healthy cells from pathological ones. However, a major limitation of atomic force microscopy (AFM)—a widely used technique for such measurements—is its low throughput and lack of standardization[1]. In this study, we optimized AFM-based mechanical measurements on cell populations and developed a novel technology that integrates cell patterning with AFM automation, significantly increasing measurement efficiency. [2]Our system enables the acquisition of mechanical data from hundreds of cells, with 956 cells analyzed in this study. For each cell, 16 force curves (FCs) were recorded, and seven key mechanical features per FC were extracted, forming a comprehensive mechanome dataset. To classify these measurements, we employed a machine learning-based approach using a fuzzy logic algorithm trained to distinguish between nonmalignant and cancerous cells. The training dataset included up to 120 cells per cell line.As a proof of concept, we first applied our method to prostate cell lines—nonmalignant RWPE-1 and cancerous PC3-GFP—before extending it to skin fibroblast lines—nonmalignant Hs 895.Sk and cancerous Hs 895.T. Despite a high degree of similarity across measurements (ranging from 79% to 100%), our method achieved a classification accuracy of 73% on a validation dataset comprising 194 cells per cell line.These results demonstrate the potential of combining AFM automation with machine learning for high-throughput mechanobiological cell classification. This approach not only enhances measurement efficiency but also provides a standardized framework for analyzing cell mechanics, paving the way for future applications in cancer diagnostics and mechanobiology research.Références : exemple de format ci-dessous[1] Thomas- -Chemin, et al. ACS Nano 19.5 (2025): 5045-5062[2] Thomas - - Chemin, et al., ACS Applied Materials and Interfaces 16.34 (2024): 44505-44517Adresse mail : [email protected]