Portail HAL UPEC
Not a member yet
55480 research outputs found
Sort by
Les discriminations à l'embauche dans l'enseignement supérieur et la recherche
No full textNous mesurons les discriminations dans l'accès à l'emploi au sein du domaine de l'enseignement supérieur et de la recherche à l'aide d'un test par correspondance, dont le principe consiste à comparer les réponses des recruteurs à des candidatures fictives semblables en tous points à l'exception d'une caractéristique dont on veut évaluer l'effet. Deux critères sont testés, le genre des candidats et leur origine géographique, dans deux professions, assistant de gestion et responsable administratif, à l'aide de deux méthodes de test, en réponse à des offres d'emplois et à l'aide de candidatures spontanées adressées à des recruteurs. Au total, l'étude exploite les résultats de 2 748 candidatures envoyées en 2021 et 2022 à des établissements d'enseignement supérieur et de recherche en France. Dans ce secteur particulier où les principes d'égalité et d'intégrité structurent largement l'activité, l'étude met en évidence une exposition aux discriminations selon l'origine Nord-Africaine d'une ampleur comparable à celle qui existe dans l'ensemble de la fonction publique. Elle souligne également la sensibilité des résultats à la couverture géographique retenue dans les tests mesurant l'intensité des discriminations
Internal structure of random polydisperse foams in steady state
No full textInternational audienceThe interplay between the foam structure and the liquid fraction, which has been established theoretically, is difficult to access experimentally due to foam aging. Here, we use small-angle neutron scattering coupled with optical imaging to investigate the structure of foams under stationary conditions where the liquid fraction remains constant over time. Through forced drainage, i.e., by injecting the foaming solution at a low flow rate above the foams, the coalescence and coarsening of polyhedral bubbles are suppressed. At the nanoscale, the thin film thickness and specific surface area of the films and Plateau borders swell and reach constant values during forced drainage. Although foam is by nature a non-equilibrium system, the low drainage flow rates considered place the structure in a steady state. Under these conditions, theoretical models based on Kelvin cells for monodisperse foams enable to predict the structure of our foams, despite being polydisperse, as long as they are in a steady state obtained in forced drainage
Don à des inconnus et diversité populationnelle : faut-il repenser la place de l'inconnu dans la communication sur le don du sang ?
No full textInternational audienceThe diversity of the population living in a Western context poses a major challenge for blood collection organizations, which must work to attract donors from all origins. Faced with an underrepresentation of ethnic minorities among donors, blood collection agencies are seeking to develop strategies to encourage these populations to give blood. Although all populations have similar blood types, people of African and Caribbean descent sometimes carry a rare phenotype that must be taken into account when transfusing other afro-caribbeans. Falling within the fields of both social marketing and ethnic marketing, this doctoral work analyzes the relationship between ethnic identity and the adoption of prosocial health behaviors, particularly blood donation. two quantitative studies based on experiments were conducted with 319 and 249 participants, using a multicultural approach (afro-caribbean vs caucasian). The results reveal that the cultural adaptation of messages calling for blood donation increases ethnic identification with the message and the intention to donate blood among afro-caribbeans, while eliciting more nuanced effects among caucasians. By linking « ethnic identity » and « blood donation », this research highlights, on a theoretical level, that ethnicity is not only a determining factor in consumer behavior, but also has a significant influence on the adoption of healthy behaviors. On a practical level, managerial recommendations are made to blood collection organizations to improve donor recruitment, to community associations wishing to increase the participation of their members, and to public authorities with a view to providing a favorable environment for the recruitment of ethnic minorities.la diversité de la population vivant dans un contexte occidental constitue un défi majeur pour les organismes de collectes du sang, qui doivent travailler à attirer des donneurs de toutes les origines. Face à une sous-représentation des minorités ethniques parmi les donneurs, les organismes de collecte cherchent notamment à développer des stratégies visant à inciter ces populations à donner leur sang. En effet, même si toutes les populations ont des groupes sanguins similaires, les populations originaires d'Afrique et des Caraïbes sont parfois porteuses d'un phénotype rare dont il faut tenir compte pour transfuser d'autres afro-caribéens. S'inscrivant à la fois dans le champ du marketing social et du marketing ethnique, cette recherche analyse les relations entre l'identité ethnique et l'adoption de comportements de santé prosociaux, en particulier le don de sang. Ainsi, deux études quantitatives à base d'expérimentations ont été menées auprès de 319 et 249 participants, selon une approche multiculturelle (afro-caribéens vs caucasiens). Les résultats révèlent que l'adaptation culturelle des messages d'appel au don de sang accroît l'identification ethnique au message ainsi que l'intention de donner son sang chez les afro-caribéens, tout en suscitant, chez les caucasiens, des effets plus nuancés. En liant « identité ethnique » et « don du sang », cette recherche met en évidence, sur le plan théorique, que l'ethnicité ne constitue pas uniquement un facteur déterminant dans les comportements de consommation, mais qu'elle exerce également une influence significative sur l'adoption de comportements de santé. Sur le plan pratique, des recommandations managériales sont formulées à destination des organismes de collectes du sang pour améliorer le recrutement de donneurs, aux associations communautaires désireuses de renforcer la participation de leurs membres, ainsi qu'aux pouvoirs publics dans une perspective d'offrir un cadre favorable au recrutement des minorités ethniques
The Vassal game engine as a support for modeling digital forces
No full textPAXsimsAs part of the EdUTeam educational and research project on game-oriented learning, conducted at IAE Paris-Est and the Institut de Recherche en Gestion (IRG, UR 2354), we are studying the different uses of wargaming in social sciences (Lépinard, 2025). This article reports on exploratory work carried out in February 2026 during a 28-hour introductory course on wargaming with 17 students enrolled in the Master 2 program in Information Systems Security Management
Functional connectivity of the amygdala subnuclei in various mood states of bipolar disorder
No full textInternational audienceAmygdala functional dysconnectivity lies at the heart of the pathophysiology of bipolar disorder (BD). Recent preclinical studies suggest that the amygdala is a heterogeneous group of nuclei, whose specific connectivity could drive positive or negative emotional valence. We investigated functional connectivity (FC) changes within these circuits emerging from each amygdala's subdivision in 127 patients with BD in different mood states and 131 healthy controls (HC), who underwent resting-state functional MRI. FC was evaluated between lateral and medial nuclei of amygdalae, and key subcortical regions of the emotion processing network: anterior and posterior parts of the hippocampus, and core and shell parts of the nucleus accumbens. FC was compared across groups, and subgroups of patients depending on their mood states, using linear mixed models. We also tested correlations between FC and depression (MADRS) and mania (YMRS) scores. We found no difference between the whole sample of BD patients vs. HC but a significant correlation between MADRS and right lateral amygdala /right anterior hippocampus, right lateral amygdala/right posterior hippocampus and right lateral amygdala/left anterior hippocampus FC (r = -0.44, r = -0.32, r = -0.27, respectively, all pFDR<0.05). Subgroup analysis revealed decreased right lateral amygdala/right anterior hippocampus and right lateral amygdala/right posterior hippocampus FC in depressed vs. non-depressed patients and increased left medial amygdala/shell part of the left nucleus accumbens FC in manic vs non-manic patients. These results demonstrate that acute mood states in BD concur with FC changes in individual nuclei of the amygdala implicated in distinct emotional valence processing. Overall, our data highlight the importance to consider the amygdala subnuclei separately when studying its FC patterns including patients in distinct homogeneous mood states
Inférence adaptative pour des réseaux de neurones profonds économes en ressources
No full textThe continuous rise of deep neural networks comes with an explosion in computational cost, latency, and energy consumption during inference. These constraints become critical in real-time computer vision scenarios, where multiple video streams must be processed in parallel on resource-limited edge platforms, as well as in the execution of large-scale models hosted in data centers.This thesis proposes a systematic approach based on early exits to dynamically adapt the effective depth of a network according to the difficulty of each instance. The thesis explores three aspects:Architecture: optimal design, placement, and sizing of exit branches that preserve feature hierarchy while minimizing parameter overhead;Training: hierarchical loss strategies, multi-level distillation, and statistical calibration of outputs aimed at balancing overall accuracy and reliability of intermediate estimators;Inference: policies based on statistically grounded confidence thresholds to decide on early stopping, and latency-energy measurement protocol on server GPU and Jetson platforms.Major contributions include: i) a systematic exploration of existing methods across the entire early exit deployment chain (architecture, training, and inference); ii) a rejection-based inference approach enabling efficient use of early exits in resource-constrained environments; iii) a specific training and inference methodology adapted to large language models (LLMs) allowing the exploitation of early exits to significantly reduce computational costs while preserving the quality of generated responses.All code, profiling scripts, and trained models are published as open-source to promote research on energy efficiency in deep learning.The results confirm that \textit{early exit} networks constitute a pragmatic solution to reconcile performance and digital sobriety, paving the way for secure and privacy-respectful deployments on the \textit{edge} and beyond.L'essor continu des réseaux de neurones profonds s'accompagne d'une explosion du coût computationnel, de la latence et de la consommation énergétique lors de l'inférence. Ces contraintes deviennent critiques dans les scénarios de vision par ordinateur temps réel, où plusieurs flux vidéo doivent être traités en parallèle sur des plateformes edge à ressources limitées, ainsi que dans l'exécution de modèles à grande échelle hébergés dans des centres de données. Cette thèse propose une approche systématique fondée sur les sorties anticipées (early exit) pour adapter dynamiquement la profondeur effective d'un réseau en fonction de la difficulté de chaque instance. Le manuscrit explore trois aspects:Architecture : conception, placement et dimensionnement optimaux des branches de sortie permettant de préserver la hiérarchie des caractéristiques tout en minimisant le surcoût paramétrique;Entraînement : stratégies de pertes hiérarchiques, distillation multi-niveaux et calibration statistique des sorties visant à équilibrer précision globale et fiabilité des estimateurs intermédiaires;Inférence : politiques basées sur des seuils de confiance statistiquement fondés pour décider de l'arrêt précoce, et protocole de mesure latence-énergie sur platesformes GPU serveur et Jetson.Les contributions majeures incluent : i) une exploration systématique des méthodes existantes à travers toute la chaîne de déploiement de l'early exit (architecture, entraînement et inférence); ii) une approche d'inférence basée sur le rejet permettant l'utilisation efficace des sorties anticipées dans des environnements à ressources contraintes; iii) une méthodologie spécifique d'entraînement et d'inférence adaptée aux modèles de langage de grande taille (LLMs) permettant d'exploiter les sorties anticipées pour réduire significativement les coûts computationnels tout en préservant la qualité des réponses générées.L'ensemble du code, des scripts de profilage et des modèles entraînés est publié en open-source afin de favoriser la recherche sur l'efficacité énergétique du deep learning.Les résultats obtenus confirment que les réseaux à early exit constituent une solution pragmatique pour concilier performances et sobriété numérique, ouvrant la voie à des déploiements sécurisés et respectueux de la vie privée sur l'edge et au-delà
Building localized and generic scenarios for sea and river dikes based on their socio-ecological footprint
No full textInternational audienceIn light of climate change and a new regulatory context, the DIGUES research program is exploring the potential futures of French dikes up to 2050. An interdisciplinary, forward-looking approach has been applied to consider the socio-ecological footprint of the ways in which dikes evolve. It brings together social and environmental sciences with stakeholder participation to build localized and generic scenarios. This approach offers an alternative to technical approaches based solely on sea or river defense, as it puts forward a method that links the physical dynamics of the environment to social and organizational issues and variables. The pathways presented (dikes maintained in their current state, reinforced, lowered, or moved back) therefore position the future of dikes in terms of continuity or transformative strategies in a context where territories take on greater responsibility in the face of climate change. In diking strategies, local public action is still driven by territorial goals and should consider more environmental dynamics. Developing territorial projects that include social and ecological elements helps to promote transformative strategies
Real-World Outcomes of First-Line Immune-Based Combination Therapies in Bone-Metastatic Clear Cell Renal Cell Carcinoma
No full textInternational audienceBone metastases affect one-third of patients with metastatic clear cell renal cell carcinoma and remain a major cause of morbidity. In this multicenter real-world study of 124 patients treated with first-line ICI-ICI or ICI-TKI combinations, both strategies showed comparable outcomes. Bone progression and skeletal events remained frequent, with bone-targeting agents independently associated with improved bone progression-free survival. Background: Bone metastases occur in one third of patients with metastatic clear cell renal cell carcinoma (mccRCC) and are associated with poor prognosis. Optimal first-line treatment for this subgroup of patients remains undefined. Methods: We conducted a multicenter retrospective study of patients with bone-metastatic (BM + ) mccRCC treated with first-line immune checkpoint inhibitor (ICI)-based combinations between January 2015 and February 2024 across 8 French centers. The primary endpoint was time to next treatment (TTNT). Secondar y endpoints included overall sur vival (OS), bone progression-free survival (bPFS), incidence of skeletal-related events (SREs) and changes in bone pain. Results: A total of 124 patients were included; 55% received ICI-ICI and 45% an ICI plus a tyrosine kinase inhibitor (TKI). Median follow-up was 36.8 months. Median TTNT was 11.7 months (95% CI, 8.38-20.0), OS 28.8 months (95% CI, 23.7-40.2) and bPFS 11.7 months (95% CI, 7.69-21.7). Median TTNT was numerically longer with ICI-TKI compared to ICI-ICI (17.9 vs. 8.5 months; P = .40), with no significant difference in OS or bPFS between treatment groups. SREs occurred in 32% of patients. Bone progression was observed in 48%, with a concomitant SRE in 54% mostly involving preexisting lesions (78%). Bone pain improved in 45% of evaluable patients. No significant differences were observed between treatment groups for SRE incidence or bone pain improvement. Hypercalcemia was associated with shorter OS and bPFS, while bone-only disease correlated with improved OS. Use of bone-targeting agents (BTAs) was independently associated with longer bPFS. Conclusion: ICI-CI and ICI-TKI combinations showed comparable outcomes in BM + mccRCC. Bone progression and SREs remained frequent and often involved preexisting bone lesions. These findings support early integration of local treatments as well as BTAs, which were independently associated with longer bPFS
Entre qualité des soins et injonctions budgétaires : les paradoxes du NPM à l’écran, en France, au Royaume-Uni et aux USA
No full textInternational audienc
Enhancing the CO2 Separation Performance of MOFs-Based Membranes: From Strategic Modifications to Computational Insights
No full textInternational audienceCarbon dioxide (CO2) separation and capture technologie have been considered as a viable strategy for mitigating CO2 emissions to reduce the greenhouse effect. Metal organic frameworks (MOFs)-based membrane separation technology has demonstrated to be an efficient and sustainable approach for CO2 separation and capture compared to the conventional methods. The development of high-performance materials and fabrication techniques, assisted with modern computational methods, has become an active research area. In this review, the state of the art and applications of MOFs-based membrane for CO2 separation technology were summarized, and the materials synthesis and modification methods reported in the last five years were comprehensively compared to evaluate the advantages and limitations in improving the permeability and selectivity of the membranes. The most recent progress of computational methods involving molecular simulations and machine learning was outlined to understand the underlying molecular mechanisms of the separation performance and high-throughput screening of materials. Finally, the challenges and prospects of the current development status of MOFs-based membranes in both experiments and data-driven methods for CO2 separation were addressed