Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Bridging Open-Source Photogrammetry : Toward Synergies Between Meshroom, MicMac and others
No full textContributors: Anthony Pamart (AP), Ewelina Rupnik(ER), Fabien Castan (FCA), Antoine Laurent (AL),Marco Gaiani (MG), Gregoire de Lillo (GDL), BenoitMaujean (BM), Florent Comte (FCO), Jean-MichaelMuller (JMM), Andrea Ballabeni (AB). CRediT (https://credit.niso.org/): Conceptualization : AP; Datacuration : AP; Formal analysis : AP; Funding acquisition: AP; Investigation : AP; Methodology : AP; Project administration: AP; Resources : AP, FCA, AL; Software: AP, ER, FCA, MG, GDL, BM, FCO, JMM, AB; Supervision: AP; Validation : AP; Visualization : AP;Writing – original draft : AP, ER, FCA, MG; Writing –review & editing : AP, ER, FCA, MG.International audienceAbstract. This paper introduces MicMacRoom, a prototype that bridges the qualitative photogrammetric workflow of MicMac with the intuitive, node-based interface of Meshroom, addressing critical limitations of open-source photogrammetry in usability, interoperability, and sustainability. Developed by a working group of the French National 3D Consortium, MicMacRoom targets Cultural Heritage applications by providing customizable workflows for key processes ; dense cloud generation, meshing, and orthophoto production. As an open-source project, MicMacRoom provides public access to its source code, documentation, and processing templates, laying the foundation for a unified and sustainable photogrammetry ecosystem. The current version of the prototype is a starting point for future developments, focusing on expanding interoperability with additional image-based modeling tools and methods. This article also presents the ongoing integration of colour calibration as a preliminary step for ensuring future compatibility with advanced multi-view photometric workflows
Proposition d’un guide réflexif pour la construction de démarche d’innovation collective dans les filières agroalimentaires: Document de travail
No full textLes sciences économiques et de gestion s’accordent à définir tout système d’innovation comme un réseau d’acteurs qui interagissent pour le développement d’une innovation grâce à la mobilisation de ressources (financières, humaines et matérielles) et au sein de règles d’action partagées. Quatre dimension clés structurent la performance de tout système d’innovation : la pertinence et cohérence des acteurs en présence ; l’intensité des interactions les liant ; les règles d’action collective (ie. régles du jeu) qui déterminent leur capacité à agir (on distingue souvent les institutions formelles et plus informelles liées aux routines et croyances) ; les ressources leur permettant d’agir (dénommés aussi infrastructures matérielles et non-matérielles). Analyser chacune de ces dimensions est reconnu comme trop complexe. C’est pourquoi il est plus simple de partir d’une catégorisation d’actions concrètement réalisées par les acteurs afin d’évaluer leur capacité d’innovation. Ce guide propose une catégorisation en 8 fonctions-clés pour évaluer et accompagner la performance des filières agroalimentaires à innover : entrepreneuriat, construction de marchés, mobilisation de ressources, extension du réseau, gouvernance et coordination, accroissement de la notoriété, développement et diffusion des connaissances, direction du changement. Le guide permet d’aborder chacune de ces fonctions au travers d’un ensemble de questions pour guider la réflexion d’une stratégie d’innovation pour tout opérateur des filières
Continuized Nesterov Momentum Achieves the Complexity without Additional Mechanisms
No full textFor first-order optimization of non-convex functions with Lipschitz continuous gradient and Hessian, the best known complexity for reaching an -approximation of a stationary point is . Existing algorithms achieving this bound are based on momentum, but are always complemented with safeguard mechanisms, such as restarts or negative-curvature exploitation steps. Whether such mechanisms are fundamentally necessary has remained an open question. Leveraging the continuized method, we show that a Nesterov momentum algorithm with stochastic parameters alone achieves the same complexity in expectation. This result holds up to a multiplicative stochastic factor with unit expectation and a restriction to a subset of the realizations, both of which are independent of the objective function. We empirically verify that these constitute mild limitations
Acting without acting: Sicilian "Mòviti fermo!" `Move still!' between human judgment andlanguage models
No full textInternational audienceNatural languages do not only describe actions and states, but also provide resources to manage actions that are anticipated, discouraged, or redirected in interaction. This talk investigates a striking case from Sicilian, in which the verb MOVE can be used in apparently paradoxical directives such as "Mòviti fermo!" (lit. ‘Move still!’), instructing the addressee not to move. Rather than analysing it as a case of semantic reversal or contronymy, we argue that this usage reflects a general mechanism by which a dynamic verb is used to constrain and reorient an expected course of action. On this view, the apparent static meaning of MOVE arises only in contexts in which an alternative free movement is pragmatically available and must be neutralised. We explore this hypothesis through a parallel evaluation involving acceptability judgments from native speakers of Sicilian and distributional representations from a BERT-based language model. The results suggest that Sicilian MOVE does not encode stasis per se but an operation of action control. We discuss the implications of this view for the semantics–pragmatics interface and the limits of current language models in capturing such interactional meanings
Revisiting Incremental Stochastic Majorization-Minimization Algorithms with Applications to Mixture of Experts
No full textProcessing high-volume, streaming data is increasingly common in modern statistics and machine learning, where batch-mode algorithms are often impractical because they require repeated passes over the full dataset. This has motivated incremental stochastic estimation methods, including incremental stochastic Expectation-Maximization (EM) procedures formulated via stochastic approximation. In this work, we study an incremental stochastic variant of the Majorization-Minimization (MM) principle that generalizes incremental stochastic EM as a special case. Our main methodological contribution is to make this framework operational for softmax-gated mixture of experts (MoE) models: we construct tractable majorizer surrogates that yield explicit incremental updates in settings where no stochastic EM algorithm is available. We complement this development with consistency guarantees by verifying conditions under which the iterates converge to a stationary point characterized by a vanishing gradient of the objective. Empirically, for softmax-gated MoE regression, the resulting incremental stochastic MM algorithm consistently outperforms widely used stochastic optimizers, including stochastic gradient descent, root mean square propagation, adaptive moment estimation, and second-order clipped stochastic optimization. These results highlight the practical value of building a principled bridge between incremental MM theory and modern softmax-gated MoE architectures, given their central role in contemporary deep neural networks for heterogeneous data modeling and scalable conditional computation. Beyond synthetic experiments, we further validate effectiveness on two real-world datasets, including a bioinformatics study of dent maize genotypes under drought stress that integrates high-dimensional proteomics with ecophysiological traits, where incremental stochastic MM yields stable gains in predictive performance
From non-specific biomarker to targeted action: transdiagnostic and sex-specific drivers of high-CRP status in severe mental illness across the FondaMental Advanced Centers of Expertise (FACE) cohorts
No full textInternational audienceBackground and objectives: Low-grade systemic inflammation contributes to the pathophysiology of severe mental illness (SMI) in a substantial subset of patients, who often experience greater disease burden and poorer treatment response. Elevated C-reactive protein (CRP), defined as CRP ≥ 3 mg/L, has been proposed to identify this group, but its non-specificity limits the biomarker's ability to guide targeted intervention. We aimed to determine the most consistent drivers of high CRP across bipolar disorder (BD), schizophrenia (SZ) and major depressive disorder (MDD), and to translate these into clinically actionable intervention targets using robust data-driven methods.Methods: We pooled and harmonised data from three large French national SMI cohorts (n = 7149: 4797 bipolar disorder, 1958 schizophrenia and 394 resistant major depression) and classified participants by CRP ≥ 3 mg/L, as well as an alternative cut-off of 5 mg/L. We applied penalised logistic regression (PLR), random forests (RF) and unsupervised clustering, using 28 biopsychosocial variables to identify robust drivers of high-CRP status. We then grouped these into actionable targets and assessed relative dominance.Results: In total, 30.16% of participants had CRP ≥ 3 mg/L. PLR identified female sex (OR [95% CI]: 1.60 [1.27, 1.93]), higher BMI (OR: 1.09 [1.07, 1.13]), current nicotine dependence (OR: 1.05 [1.02, 1.09]), lower HDL cholesterol (OR: 0.57 [0.44, 0.73]) and smoking (ex-smoker status OR: 0.84 [0.66, 0.98]) as consistent drivers. RF highlighted a similar set of key drivers, also including waist circumference, triglycerides and cardiovascular comorbidities. Clustering of the high-CRP group was almost entirely driven by smoking status and nicotine dependence. When grouped into actionable targets, the identified drivers accounted for 16% of variance in CRP status, with obesity emerging as most dominant contributor. This pattern was most pronounced in females; in males it was more diffuse, with a more prominent role for smoking.Conclusions: We propose a decision tree framework where CRP can serve as a first-line screening marker for inflammation in SMI, with subsequent steps focusing on the main contributing factors to guide targeted interventions. Priority should be given to targeting obesity and metabolic dysregulation. Among females, hyperuricemia represents the next most appropriate target, whereas in males, smoking warrants greater attention. This stepwise approach provides a route from a nonspecific biomarker to targeted treatment strategies and should be validated in prospective studies
Structurer les interactions entre recherche et gestion de la biodiversité : Retour d’expérience du GT “Biodiversité” (2021-2024)
No full textThe common objectives of MSH and BiodivOc in terms of multidisciplinarity and inter-community synergies are at the origin of this initiative. Together, they have created the Working Group (WG) ‘Biodiversity’: a tool for dialog and co-production of knowledge composed of members of BiodivOc, MSH Sud (Trait d'union scheme), the Regional Biodiversity Agency (ARB), Man And the Biosphere (MAB France), and the Natural History Museum of Toulouse. This WG has tested a new interface that establishes a concrete and dynamic link, with a view to sustainability, between research stakeholders and biodiversity management stakeholders in Occitanie (parks, agencies and associations). The main objective was to identify the needs of the various communities (research, expertise, methodological support, experimental sites, etc.) and to provide an appropriate, concrete and applicable response by bringing together research and field stakeholders (provision of existing knowledge, scientific support, etc.) and/or developing collaborative projects (research projects, student internships, etc.). The aim was to establish lasting synergies between research, biodiversity management stakeholders and civil society in Occitanie, and to develop/strengthen the biodiversity system within Trait d'union (and other science shops or similar intermediation systems that may be created in Occitanie) by offering a privileged space for mutual understanding and resources capable of guaranteeing the conditions for the co-construction of research in the field of ‘Biodiversity’, in accordance with the rules of scientific ethics and in a spirit of fairness between academic partners – from all disciplines – and non-academic partners.Les objectifs communs de la MSH et de BiodivOc en matière de pluridisciplinarité et de synergies inter-communautés sont à l’origine de cette initiative. Ensemble, ils ont créé le Groupe de travail (GT) « Biodiversité » : un outil de concertation et de co-production des connaissances composé de membres de BiodivOc, de la MSH Sud (dispositif Trait d’union), de l’Agence régionale de biodiversité (ARB), de Man And the Biosphere (MAB France), et du Muséum d’Histoire Naturelle de Toulouse. Ce GT a expérimenté une interface nouvelle permettant d’établir un lien concret et dynamique, dans un objectif de pérennisation, entre acteurs de la recherche et acteurs de la gestion de la biodiversité en Occitanie (parcs, agences et associations). L’objectif principal était d’identifier les besoins des différentes communautés (recherche, expertise, appui méthodologique, terrains d’expérimentation, etc.) et d’apporter une réponse adaptée, concrète et applicable à travers un rapprochement des interlocuteurs de la recherche et du terrain (mise à disposition de connaissances existantes, appui scientifique, etc.) et/ou le développement de projets collaboratifs (projets de recherche, stages étudiants, etc.). Ce faisant, l’ambition était de structurer de manière pérenne les synergies entre recherche, acteurs de la gestion de la biodiversité et société civile en Occitanie, et de développer/renforcer le dispositif biodiversité au sein de Trait d’union (et autres boutiques des sciences ou dispositifs d’intermédiation similaires qui viendraient à se créer en Occitanie) en proposant un espace privilégié d’interconnaissance et de ressources à même de garantir les conditions la co-construction de recherches dans le domaine « Biodiversité », dans le respect des règles de la déontologie scientifique et dans un esprit d’équité entre partenaires académiques – de toutes disciplines – et non-académiques
Triangulation et intégration : capabilités, compétences à s’orienter et représentations sociales pour penser les inégalités
No full textThis article proposes a model of the orientation process conceived as a continuum between a « state-oriented » individual (defined by their attributes and subject to determinism) and an « action-oriented » individual (taking concrete action on their career path). We argue that this transition to action results from the dynamic interaction of three constituent concepts: orientation skills (personal resources), orientation capability (real freedoms offered by the environment), and social representations (cognitive and normative filters). The uniqueness of our analytical framework lies in the central place it gives to volition, defined as the dynamic process of commitment and concrete will to act. We argue that volition is the integrating and triggering mechanism that activates skills, allows individuals to seize opportunities in an enabling environment, and negotiates with social representations, whether limiting or facilitating. By articulating the contributions of Bangali (orientation skills), Sen (capability), Moscovici (representations), and action theories (volition, Broonen), this model offers a tool for analyzing obstacles and levers in orientation pathways.Cet article propose une modélisation du processus d'orientation conçu comme un continuum entre une posture d'individu « orienté état » (défini par ses attributs et subissant des déterminismes) et d'individu « orienté action » (agissant concrètement sur son parcours). Nous défendons l'idée que ce passage à l'action résulte de l'interaction dynamique de trois concepts constitutifs : les compétences à s'orienter (ressources personnelles), la capabilité en orientation (libertés réelles offertes par l'environnement) et les représentations sociales (filtres cognitifs et normatifs). La singularité de notre cadre analytique réside dans la place centrale qu'il accorde à la volition, définie comme le processus dynamique de la force d'engagement et de la volonté concrète d'agir. Nous argumentons que la volition constitue le mécanisme intégrateur et déclencheur qui active les compétences, permet de saisir les opportunités de l'environnement capacitant et négocie avec les représentations sociales, qu'elles soient limitantes ou facilitatrices. En articulant ainsi les apports de Bangali (compétence à s'orienter), de Sen (capabilité), de Moscovici (représentations) et des théories de l'action (volition, Broonen), ce modèle offre un outil pour analyser les blocages et les leviers dans les parcours d'orientation
Global Sensitivity Analysis for Engineering Design Based on Individual Conditional Expectations
No full textPublished in Aerospace Science and Technology, 2026International audienceExplainable machine learning techniques have gained increasing attention in engineering applications, especially in aerospace design and analysis, where understanding how input variables influence predictive models is essential. Partial Dependence Plots (PDPs) are widely used for interpreting black-box models by showing the average effect of an input variable on the prediction. However, their global sensitivity metric can be misleading when strong interactions are present, as averaging tends to obscure interaction effects. To address this limitation, we propose a global sensitivity metric based on Individual Conditional Expectation (ICE) curves. The method computes the expected feature importance across ICE curves, along with their standard deviation, to more effectively capture the influence of interactions. The proposed metrics are model-agnostic and can be applied to any predictive model, including but not limited to surrogate models. Furthermore, we provide a mathematical proof demonstrating that the PDP-based sensitivity is a lower bound of the proposed ICE-based metric under additive and multiplicative separability. In addition, we introduce an ICE-based correlation value to quantify how interactions modify the relationship between inputs and the output. Comparative evaluations were performed on three cases: a 5-variable analytical function, a 5-variable wind-turbine fatigue problem, and a 9-variable airfoil aerodynamics case, where ICE-based sensitivity was benchmarked against PDP, SHapley Additive exPlanations (SHAP), and Sobol' indices. The results show that ICE-based feature importance provides richer insights than the traditional PDP-based approach, while visual interpretations from PDP, ICE, and SHAP complement one another by offering multiple perspectives