Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Two-level overlapping additive Schwarz preconditioner for training scientific machine learning applications
No full textInternational audienceWe introduce a novel two-level overlapping additive Schwarz preconditioner for accelerating the training of scientific machine learning applications. The design of the proposed preconditioner is motivated by the nonlinear two-level overlapping additive Schwarz preconditioner. The neural network parameters are decomposed into groups (subdomains) with overlapping regions. In addition, the network’s feed-forward structure is indirectly imposed through a novel subdomain-wise synchronization strategy and a coarse-level training step. Through a series of numerical experiments, which consider physicsinformed neural networks and operator learning approaches, we demonstrate that the proposed two-level preconditioner significantly speeds up the convergence of the standard (LBFGS) optimizer while also yielding more accurate machine learning models. Moreover, the devised preconditioner is designed to take advantage of model-parallel computations, which can further reduce the training tim
Diffusive gradient in thin film for ultra-trace methylmercury measurements in the coastal and open sea
No full textInternational audienceMonomethylmercury (MMHg) is a potent neurotoxin causing neurodevelopmental delays and cardiovascular and immunological issues. Human exposure primarily occurs through seafood consumption due to MMHg bioaccumulation and biomagnification from seawater into marine organisms. Determining MMHg in seawater at ultratrace concentrations poses logistical and analytical challenges. Diffusive Gradient in Thin-film (DGT) samplers represent a promising solution, which captures time-averaged concentrations by preconcentrating in situ MMHg over a defined exposure time. DGT manufactured with 3-mercaptopropyl-functionalized silica (3MFS) in agarose and polyacrylamide gels were tested and compared for the determination of MMHg present in open ocean and coastal waters. Different elution methods using acidic thiourea were tested to reach precise, accurate and quantitative elution of MMHg from the binding gel. We found that polyacrylamide-3MFS binding gels display a higher elution efficiency (94 ± 3 %), precision and better handling compared to agarose-3MFS gels (41 ± 6 %). A unique mooring line installed in the South Western Tropical Pacific Ocean, provided monthly DGT-MMHg concentrations over a year showing potential seasonal differences in MMHg concentrations ranging between 18 and 106 fM. DGT were also deployed in shallow Peruvian coastal waters, exhibiting higher MMHg concentrations (170 ± 97, n = 26) with typical benthopelagic gradients. DGT-MMHg concentrations were in good agreement with discrete water samples analyzed by reference methods using isotope dilution. DGTs offer complementary advantages over oceanographic cruises, notably in situ preconcentration, low blanks, minimal logistical requirements and cost-effectiveness. DGTs represent a valuable tool for studying the marine MMHg cycle for evaluating the implementation of the Minamata Convention
THE EVOLUTION OF POINTWISE STATISTICS IN HYPERBOLIC EQUATIONS WITH RANDOM DATA
No full textInternational audienceWe consider one-dimensional hyperbolic PDEs, linear and nonlinear, with random initial data. Our focus is the pointwise statistics, i.e., the probability measure of the solution at any fixed point in space and time. For linear hyperbolic equations, the probability density function (PDF) of these statistics satisfies the same linear PDE. For nonlinear hyperbolic PDEs, we derive a linear transport equation for the cumulative distribution function (CDF) and a nonlocal linear PDE for the PDF. Both results are valid only as long as no shocks have formed, a limitation which is inherent to the problem, as demonstrated by a counterexample. For systems of linear hyperbolic equations, we introduce the multi-point statistics and derive their evolution equations. In all of the settings we consider, the resulting PDEs for the statistics are of practical significance: they enable efficient evaluation of the random dynamics, without requiring an ensemble of solutions of the underlying PDE, and their cost is not affected by the dimension of the random parameter space. Additionally, the evolution equations for the statistics lead to a priori statistical error bounds for Monte Carlo methods (in particular, Kernel Density Estimators) when applied to hyperbolic PDEs with random data
Se (trans)former auprès de la nature. Une lecture des Conférences sur l'avenir des établissements d'enseignement
No full textInternational audienc
D’une culture humaniste et littéraire « classique » à une culture de l’entreprise : la redéfinition du curriculum d’une classe préparatoire littéraire B/L
No full textInternational audienc
I Feel Competent, Therefore I Am: Self-Concept and Skill Interact at Different Speeds
No full textDo perceptions about one’s competence shape learning, or are they simply reflections of actual skills? This study revisits this longstanding question by replicating and extending preliminary findings by Marsh et al. (2024) on the temporal dynamics linking students’ academic self-concept (i.e., their perceived academic competence) and their academic skills in mathematics and French (language arts). Using longitudinal data from a large-scale field study (N > 9000 students, 3 measurement points), we tested how academic self-concept and skills relate to each other over time. Consistent with Marsh et al., results revealed a consistent temporal asymmetry: Academic skills predicted concurrent changes in self-concept within the same semester (contemporaneous effects), whereas self-concept predicted changes in academic skills across semesters (lagged effects). These findings were robust to several stress tests, including measurement error, unmeasured confounding, and competing models of change. Together, the results are consistent with a renewed theory of learning behavior, in which perceived competence and skills influence each other at different speeds. This temporal asymmetry helps integrate short-term and long-term cognitive-motivational processes in theories of learning behavior. It also underscores the importance of aligning intervention strategies and model specifications with the timescales of the underlying psychological processes, with implications for both fundamental and intervention research
Linking neuroinflammation and functional connectivity disruption in coma: a graph-theoretical study
No full textInternational audienceComa causes persistent cognitive and behavioral deficits. Identifying recovery mechanisms is essential for outcome biomarkers and requires multimodal integration across imaging, modeling, and clinical practice. TSPO-PET imaging reveals neuroinflammation in key resting-state networks of coma patients, and fMRI graph analyses demonstrate functional network reorganization, yet their relationship remains unclear. To address this, we developed new modeling strategies to examine associations between fMRI graph structure and TSPO-based graphs
Setting a research agenda for the assessment and treatment of aphasia in minority languages
No full textInternational audienceThe aim of this position article is to establish the state of affairs in aphasia assessment and treatment in individuals who speak minority languages. This article reports on recommendations from a panel of experts working with individuals with aphasia in a variety of languages to develop a research agenda for aphasia assessment and treatment in minority languages. Members of Working Group 2 (Aphasia Assessment and Outcomes) of the Collaboration of Aphasia Trialists (CATs) were invited to respond to a short online agenda-setting questionnaire and to discuss issues regarding this topic. The panel of experts then refined the responses and recommendations into future research themes and objectives. Seven priority themes were identified: Definitions, Tools, Research Practices, Treatment, Speech and language pathology (SLP) Training, Societal Impact, and Norms. In the EU alone, about 60 minority/regional languages are spoken by around 40 million people. Considering increasing caseloads and a lack of clinical tools for speakers of minority languages, this research agenda has an important impact for future research and clinical advancements
Emergent Loewner Dynamics in Slime Mold Growth
No full textGrowth fronts of slime molds are characterized through a direct geometric analysis based on Loewner evolutions, using experimentally acquired time-resolved images. The associated Loewner driving functions reconstructed from expanding pseudopod boundaries display statistical properties consistent with Gaussian-like behavior.A geometric estimate of the diffusivity parameter κ is inferred from fractal scaling, while Brownian diagnostics are assessed on the reconstructed driving signal.These findings show that the boundaries of a growing living organism display statistical and geometric properties consistent with emergent Loewner dynamics over experimentally accessible scales. This study establishes a quantitative framework for analyzing biological growth interfaces and suggests new connections between morphogenesis, stochastic geometry, and network reorganization under varying environmental conditions. We provide, to our knowledge, the first explicit reconstruction of a Loewner driving function from a living growth interface, revealing an emergent Brownian-like conformal growth regime at expanding fronts