Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Enregistrement et préservation du paysage sonore de la nation en guerre dans le cinéma officiel britannique de la Seconde Guerre mondiale
No full textInternational audienc
Modélisation hiérarchique et optimisation d'architectures de systèmes : Revue comparative et cadre unifié
No full textInternational audienceSimulation-based problems involving mixed-variable inputs frequently feature domains that are hierarchical, conditional, heterogeneous, or tree-structured. These characteristics pose challenges for data representation, modeling, and optimization. This paper reviews extensive literature on these structured input spaces and proposes a unified framework that generalizes existing approaches. In this framework, input variables may be continuous, integer, or categorical. A variable is described as meta if its value governs the presence of other decreed variables, enabling the modeling of conditional and hierarchical structures. We further introduce the concept of partially-decreed variables, whose activation depends on contextual conditions. To capture these inter-variable hierarchical relationships, we introduce design space graphs, combining principles from feature modeling and graph theory. This allows the definition of general hierarchical domains suitable for describing complex system architectures. The framework supports the use of surrogate models over such domains and integrates hierarchical kernels and distances for efficient modeling and optimization. The proposed methods are implemented in the open-source Surrogate Modeling Toolbox (SMT 2.0), and their capabilities are demonstrated through applications in Bayesian optimization for complex system design, including a case study in green aircraft architecture.Les problèmes basés sur la simulation à variables mixtes présentent fréquemment des domaines hiérarchiques, conditionnels, hétérogènes ou arborescents, ce qui complique la représentation des données, la modélisation et l’optimisation. Cet article dresse un panorama exhaustif de la littérature sur ces espaces d’entrée structurés, puis propose un cadre unifié généralisant les approches existantes. Dans ce cadre, les variables d’entrée, continues, entières ou catégorielles, peuvent être qualifiées de « méta » lorsqu’elles régissent la présence d’autres variables décrétées, autorisant ainsi la modélisation de dépendances conditionnelles et hiérarchiques. Nous introduisons également les « variables partiellement décrétées », dont l’activation dépend de contextes spécifiques. Pour formaliser ces relations, nous définissons des graphes d’espace de conception (design space graphs) combinant notions de feature modeling et théorie des graphes, capables de décrire des architectures systèmes complexes. Le cadre prend en charge des modèles de substitution (surrogate models) adaptés à ces domaines, et intègre des distances hiérarchiques ainsi que des noyaux dédiés pour optimiser efficacité et précision. Les méthodes proposées sont implémentées dans la boîte à outils libre Surrogate Modeling Toolbox (SMT 2.0) et validées via des applications d’optimisation bayésienne, notamment une étude de cas portant sur l’architecture d’avions écologiques
Certified Enumeration of AI Explanations: A Focus on Monotonic Classifiers
No full textInternational audienceThe theory of minimal explanations offers a rigorous, model-based solution to the problem of producing explanations for the decisions of AI models. In some high-stakes contexts, there is a need to generate all possible explanations for a particular decision using certified programs, whose output can be trusted. We used the proof assistant Coq to certify a recently proposed algorithm for the enumeration of explanations in the case of monotonic classifiers. Our experimental results on the extracted code showcase the scalability of this approach, underscoring its potential for improving trust and reliability in AI systems
Habitat quality assessment of temperate forest ecosystems: An airborne LiDAR-based approach to predict the Index of Biodiversity Potential (IBP) at large scale
No full textInternational audienceThe Index of Biodiversity Potential (IBP) assesses the forest stand’s capacity to host species based on 10 structural, compositional, and environmental factors. Widely used by French forest managers, its reliance on in-situ surveys limits large-scale applications. While LiDAR-derived metrics can finely describe forest structure, their relationship with the IBP remains unexplored.We aimed to study these relationships with the IBP management factors, some of which reflect forest structure such as the number of large trees and vertical strata. Using a dataset of 1536 IBP plots across France, we computed LiDAR-derived structural metrics along with other variables (e.g., topographic, spectral). We then analysed their statistical relationships with the IBP factors, and calibrated predictive models using both regression and classification machine learning algorithms. Finally, we mapped the IBP management score for the first time over a 890 km area within the forests of the Ariege Pyrenees Regional Natural Park (France).The results revealed strong correlations between the IBP management score, its factors, and remote sensing metrics. LiDAR-derived metrics describing canopy height and vertical complexity were particularly important for prediction, as well as biomass and topographic metrics. Our best model, with an RMSE of 5.24 ± 0.63, predicts IBP within 5 points—a threshold beyond which variations reflect actual changes in species richness within the forest stand.These findings emphasise the relevance of remote sensing data, in particular LiDAR, for describing structural field metrics. They demonstrate that remote sensing offers a viable approach for large-scale IBP assessment
The compressible Euler system with damping in hybrid Besov spaces: global well-posedness and relaxation limit
No full textWe investigate the global well-posedness of the compressible Euler system with damping in R^d (d ≥ 1) and its relaxation limit toward the porous medium equation. In [12], the first author and Danchin studied these two problems in hybrid Besov spaces, where the high-frequency components of the solution are bounded in L2 -based norms, while the lowfrequency components are controlled in L p -based norms with p ∈ [2, max{4, 2d/(d -2)}]. Motivated by the observation that the limit system is well-posed in Lp -based spaces for p ∈ [2, ∞), we extend the low-frequency analysis to this full range, thereby providing a more unified framework for studying such relaxation limits.The core of our proof consists in establishing refined product and commutator estimates describing sharply the interactions between the high, medium, and low-frequency regimes. A key observation underlying our analysis is that the product of two functions localized at low frequencies generates only interactions between low and medium frequencies, never purely high-frequency ones. Consequently, for a suitable choice of frequency threshold, the high-frequency projection of the product of two functions localized low frequencies vanishes.</p
A Photometric Approach to Digital Image Correlation with a Super-Resolved Digital Twin (SR-PhDIC)
No full textInternational audienceDigital Image Correlation typically involves deforming a pixelated image in order to compare its grey levels with those of another image. To achieve sub-pixel accuracy, grey-level interpolation is required. However, this interpolation is non-physical and introduces biases that become particularly detrimental under finite strains. In this work, we propose an alternative photometric approach that entirely avoids interpolation, grounded in a rigorous formulation of the direct image formation problem. The inverse problem is then posed as the joint estimation of a super-resolved digital twin-representing the scene and sensor characteristics-and the displacement fields. Both are estimated by minimising a single cost function that compares all available real images to their synthetic counterparts generated through a physically based rendering model. This minimisation is performed using an efficient alternating minimisation scheme. Several two-dimensional test cases are analysed, demonstrating that the proposed method is effectively unbiased and exhibits significantly lower uncertainties than state-of-the-art DIC techniques
Stigma in cancer: Comparing community and health professionals’ acceptability of smoking and alcohol consumption
No full textInternational audienceColorectal and lung cancers are among the most common and deadly worldwide, often carrying significant stigma—especially when linked to preventable behaviors like drinking and smoking. Alcohol increases colorectal cancer risk, while smoking is the primary cause of lung cancer, leading to perceptions of self-infliction that can worsen distress, delay diagnosis, and hinder treatment. This study is the first to compare the acceptability of smoking versus drinking in cancer contexts. Using a factorial design, 132 community members and 126 healthcare professionals evaluated 72 scenarios varying by cancer type, behavior levels, diagnosis stage, post-diagnosis behavior, and activity level. Overall, drinking was viewed as more acceptable than smoking, especially among those physically active or who quit post-diagnosis. Healthcare professionals were slightly more accepting than the general public. Notably, stigma around smoking lessened at advanced cancer stages. Findings suggest the need for stigma-reducing interventions and more compassionate, nonjudgmental care in oncology settings
Oil Cycle Dynamics and Future Oil Price Scenarios
No full textNational audienceWe review the oil cycle and evoke future scenarios as to how it will end
La ZA PYGAR : des Pyrénées aux plaines agricoles, un territoire drainé par la Garonne
No full textChapitre 14.International audienc
