Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
"La recherche de soi: la pratique du Journal en philosophie. Simone Weil"
No full textInternational audienc
Drivers of Vegetation Structure Differ Between Proposed Natural Reference Conditions for Temperate Europe
No full textInternational audienceAim: Pre-degradation baseline conditions (references) provide crucial context for restoration actions. Here, we compare vegetation structure and its driving processes across the main pre-agricultural references discussed for temperate Europe: the Last Interglacial and the early-mid Holocene-before and after the arrival of Homo sapiens, respectively.Location: Temperate Europe.Time Period: The first ~4000-6000 years, excluding the initial early-successional phases, of the Last Interglacial (PAAZ III) and Holocene (8700-5700 BP).Major Taxa Studied: Plants.Methods: We use large datasets of pollen-based vegetation reconstructions (REVEALS) to compare open vegetation, light woodland and closed forest between the two periods. We use Random Forest modelling and downscaled climate data to assess whether climate-vegetation relations were consistent between periods, as expected if they reflected direct climatic effects on vegetation.Results: Open vegetation was slightly higher in the early-mid Holocene than in the Last Interglacial, averaging 20% versus 16% in paired grid cells, respectively. In contrast, light woodland cover was lower in the early-mid Holocene, with mean values of 49% compared to 57% in paired cells. The combined open vegetation and light woodland cover was high in both periods, averaging 73% in the Last Interglacial and 69% in the early-mid Holocene. Closed forest cover was similar across both periods (Holocene = 24%; Last Interglacial = 23%). Notably, openness -climate relations from the early-mid Holocene cannot predict open vegetation in the Last Interglacial.Main Conclusions: These findings suggest that vegetation in the early-mid Holocene and Last Interglacial was affected by persistent, substantial disturbances, which were not controlled by direct climate effects, and that the main drivers differed between the periods, with the rich megafauna of the Last Interglacial and Mesolithic people as the primary candidates. Our findings support that early-mid Holocene ecosystems were already strongly shaped by Homo sapiens and differed from earlier temperate ecosystems
‘Madame née Secret’: Reversing the gendered norms of the French language in Jean Genet’s Notre-Dame-des-Fleurs
No full textInternational audienc
Digital twins for military aircraft: A machine learning approach for monitoring structural aging
No full textInternational audienceUnlike civil aviation, where regular maintenance schedules are feasible, military aircraft are subjected to highly variable flight conditions based on mission requirements. This makes real- time assessment of structural fatigue critical for safety. Traditional approaches rely on physics-based models to predict mechanical strains, fatigue, and aging from input flight control data. While these models provide significant insights, they may not fully capture the complexities of real-world conditions and can benefit from refinement. Our research seeks to enhance these models by using machine learning to create digital twins of strain gauges for military aircraft, capable of predicting structural strains and aging based on input-output flight data.A part of these flight data, such as altitude, Mach number or others, are routinely measured during flight missions. In this work, structural strains are additionally recorded at critical points on one instrumented aircraft using strain gauges. The objective is then to develop a robust machine learning framework that simulates the behavior of critical aircraft strain gauges under varying operational conditions. In other words, using flight parameters as inputs, we aim to predict the strains experienced at specific points on the aircraft. This predictive capability can improve the planning of maintenance activities, guaranteeing maintenance in operational condition, and therefore enhancing flight safety.A key step in our project deals with the exploration of the high-dimensional flight data using autoencoders and other techniques, in order to capture complex relationships between flight parameters, reduce the dimensionality of the data, and group similar configurations together. This reduction is essential for improving the efficiency and accuracy of subsequent regression models, particularly in the context of semi-supervised learning, where we leverage both the reconstruction of input data and the prediction of structural strains to compensate for the lack of labeled data. In this context, we conduct a comparative analysis of various regression models, including tree-based algorithms, deep learning models, and semi-supervised approaches.Looking ahead, several key avenues are identified. One important perspective is the incorporation of physics-based insights into the machine learning framework, creating a hybrid model that leverages both data-driven predictions and physical laws. This integration would improve the model’s accuracy and its reliability. Additionally, the explainability of the data driven models, using techniques such as LIME (Local Interpretable Model-agnostic Explanations) or GEMS-AI, is crucial. By ensuring transparency in the predictions, we can provide users with the necessary insights to make informed decisions about aircraft maintenance
FEMDA: Un framework unifié pour l'analyse discriminante
No full textInternational audienceAlthough linear and quadratic discriminant analysis are widely recognized classical methods, they can encounter significant challenges when dealing with non-Gaussian distributions or contaminated datasets. This is primarily due to their reliance on the Gaussian assumption, which lacks robustness. We first explain and review the classical methods to address this limitation and then present a novel approach that overcomes these issues. In this new approach, the model considered is an arbitrary Elliptically Symmetrical (ES) distribution per cluster with its own arbitrary scale parameter. This flexible model allows for potentially diverse and independent samples that may not follow identical distributions. By deriving a new decision rule, we demonstrate that maximum-likelihood parameter estimation and classification are simple, efficient, and robust compared to state-of-the-art methods
Transbordamentos: a feminidade equívoca de Anne Carson
No full textInternational audienceThis article examines the concept of the feminine in the work of Anne Carson, focusing on three of her essays—two academic and a literary one. Our aim is to investigate how Carson constructs a definition of the feminine in her analysis of the Ancient Greekimaginary of women, in order to show that “woman” appears there as an essentially equivocal figure. This ambiguity arises from the equally equivocal notion of matter that underlies it. We will analyze how Carson explores this ambivalence by playing with the inversion of gender positions in her literary and “auto-theoretical” essay 'The Anthropology of Water, and we argue that this ambiguity of the feminine is deeply linked to the female author’s unavoidably paradoxical position (both as an academic and as a poet), for she is situated simultaneously within and outside the logos.Este artigo examina o conceito de feminino na obra de Anne Carson, com foco em três de seus ensaios — dois acadêmicos e um literário. Nosso objetivo é investigar como Carson constrói uma definição do feminino em sua análise do imaginário da mulher na Antiguidade grega, mostrando que “a mulher” aparece ali como uma figura essencialmente equívoca, o que se deve ao conceito também equívoco de matéria que a fundamenta. Analisaremos como Carson explora essa ambiguidade ao jogar com a inversão das posições de gênero em seu ensaio literário e autoteórico “The Anthropology of Water”. Finalmente, argumentaremos que essa mesma equivocidade do feminino está profundamente relacionada à posição paradoxal da autora enquanto mulher (acadêmica e poeta), situada ao mesmo tempo dentro e fora do logos
Gradient correlation is a key ingredient to accelerate SGD with momentum
No full textInternational audienceEmpirically, it has been observed that adding momentum to Stochastic Gradient Descent (SGD) accelerates the convergence of the algorithm. However, the literature has been rather pessimistic, even in the case of convex functions, about the possibility of theoretically proving this observation. We investigate the possibility of obtaining accelerated convergence of the Stochastic Nesterov Accelerated Gradient (SNAG), a momentum-based version of SGD, when minimizing a sum of functions in a convex setting. We demonstrate that the average correlation between gradients allows to verify the strong growth condition, which is the key ingredient to obtain acceleration with SNAG. Numerical experiments, both in linear regression and deep neural network optimization, confirm in practice our theoretical results
Solving moment and polynomial optimization problems on Sobolev spaces
No full textUsing standard tools of harmonic analysis, we state and solve the problem of moments for non-negative measures supported on the unit ball of a Sobolev space of multivariate periodic trigonometric functions. We describe outer and inner semidefinite approximations of the cone of Sobolev moments. They are the basic components of an infinite-dimensional moment-sums of squares hierarchy, allowing to numerically solve non-convex polynomial optimization problems on infinite-dimensional Sobolev spaces with global convergence guarantee
Shining Light on Dark Matter: Advancing Functional Analysis of Obsidian Tools with Confocal Scanning Microscopy
No full textInternational audienceOver the past decade, confocal microscopy has increasingly been employed to examine changes in stone tool surfaces and has proven to be an accurate technique for quantifying use-wear texture. Promising results have emerged from characterizing Polish formation on experimental and archaeological flint tools. Recent studies also highlighted the potential of confocal microscopy for analyzing tools made from reflective materials, such as quartzite. In this paper, we investigate the capability of confocal microscopy to discriminate use-wear on obsidian quantitatively. We examine whether confocal microscopy and 3D texture analysis can correctly classify several worked materials that are challenging to differentiate using the optical standard method of use-wear analysis. For cutting activities, we include butchery, de-skinning a fresh hide from grease and meaty tissues, cutting tanned leather, harvesting domestic ripe cereals, harvesting semi-green wild cereals, and sawing wet limestone. As for scraping activities, we explore discriminating differences among tools used for working dry hide, dry antler, soaked antler, fresh bone, softwood, fresh reeds, and wet limestone. Our results demonstrate that these worked materials can be confidently identified in experimental tools. While other relevant factors affecting use-wear texture, such as the intensity of use or post-depositional alterations, need to be controlled before employing the method on archaeological materials, our research suggests that the quantitative approach can enhance the standard method of use-wear analysis, providing unprecedented precision for identifying worked materials in obsidian tools
