Archives ouvertes de Paris-Saclay
Not a member yet
272730 research outputs found
Sort by
Failure mechanisms of GaN HEMTs in single event destructive short-circuit at different VDS voltage levels
No full textInternational audienceIn this work, the short-circuit (SC) robustness of 650 V normally-off SP-GaN HEMTs is investigated. The devices under test (DUTs) were subjected to single-event destructive SCs at different drain-source voltage levels. Overall, the DUTs exhibited a significant withstand time of several hundred microseconds. The experimental results first reveal the absence of a critical SC energy threshold, indicating that energy alone does not govern the failure mechanism. Instead, device failure is found to be more thermally driven, occurring once the junction temperature exceeds a specific limit. To support this conclusion, the junction temperature (Tj) evolution during the SC event was estimated using a highly simplified thermal model with a uniform heat dissipation based on the finite-element method (FEM), implemented in ANSYS APDL and calibrated with device geometries and material parameters extracted from its construction analysis. The simulation results show that all DUTs reached nearly the same Tj threshold at the instant of failure, regardless of the applied VDS. However, due to the simplifying assumptions underlying the FEM simulations presented in this study, the estimated temperatures should be regarded as indicative rather than exact
Psychological drivers of customer citizenship buying behaviour in green cosmetics: Evidence from a mixed-methods approach
No full textInternational audienceOver the past decade, the cosmetic industry has increasingly adopted green practices in response to customers' growing environmental concerns and the need to support environmental sustainability. Although there is extensive literature on green cosmetics buying behaviour, understanding how customers' extra-role behaviours respond to these green practices is still a challenging and understudied topic. This study, based on social exchange theory, investigates how perceptions of green cosmetic product attributes influence customer citizenship behaviour intention (Green CBBi) (Hassan et al., 2025). It explicitly investigates whether green product values and green credibility stimulated by green cosmetic product attributes bolster green brand connection, which in turn fosters Green CBBi and the moderating mechanism of brand neophobia in this association. The hypothesised propositions were tested using a mixed-methods approach. The present study employed a grounded theory approach to identify themes and variables using in-depth interviews. The conceptual framework was then empirically tested using Hayes' PROCESS macro with Model 80 to test hypothesised mediation effects and Model 15 for moderated mediation association with the outcome. The findings demonstrate that green cosmetic product attributes significantly affect intentions related to GCCBi, green perceived value, and green credibility. However, green perceived value does not mediate brand connection, whereas green credibility shows a negative indirect influence on brand connection, indicating a non-linear cognitive-affective process. These findings suggest that green branding strategies should achieve a balance between credibility-driven sustainability assertions and emotionally engaging, value-aligned engagement to effectively stimulate consumer behaviour
Design Considerations for Visualization Transitions of 3D Spatial Data in Hybrid AR‐Desktop Environments
No full textInternational audienceWe present design considerations for animated transitions of the appearance of 3D spatial datasets in a hybrid Augmented Reality‐desktop context. Such hybrid interfaces combine both traditional and immersive displays to facilitate the exploration of 2D and 3D data representations in the environment in which they are best displayed. One key aspect is to introduce suitable transitional animations that change between the different dimensionalities to illustrate the connection between the different representations and to reduce the potential cognitive load on the user. The specific transitions to be used depend on data type, needs of the application domain, and other factors. We summarize these as design considerations to simplify the decision‐making process and provide inspiration for future designs. We apply our concept to three case studies: sparse 3D point data, MRI scan data, and molecular data. Finally, we give some practical guidance for the prescriptive use of our design considerations
Multimodal frailty detection in primary care using a portable sensor-based platform: exploratory results
No full textInternational audienceAbstract Introduction: Frailty reflects age-related decline across multiple physiological systems, reducing resilience and increasing risks of falls, hospitalization, disability, and mortality. Scalable approaches are needed to identify pre-frailty earlier in community-dwelling older adults and enable timely prevention in primary care. Objective: To develop and evaluate a multivariable sensor-based framework for early frailty detection using standardized gait and balance assessments in general practice. Methods: We conducted a prospective cohort study (2021–2024) in southern Réunion Island among retired adults aged ≥65 years recruited in primary care. The protocol included: (1) baseline general practitioner (GP) assessment with expert frailty rating, Fried phenotype, and WHO ICOPE Step 1; (2) telephone assessment of mental health, self-rated health, and quality of life; (3) outpatient instrumented evaluation combining IMU-based gait analysis, force-platform posturography, grip strength, and ICOPE Step 2 measures; (4) monthly falls surveillance over 6 months; and (5) repeat instrumented gait and balance assessment at 6 months. Correlation analyses and machine-learning models examined relationships between frailty measures and the discriminative value of sensor-derived and multimodal predictors. Results: Among 145 participants (mean age 71±5 years), 98.5% had impairment in at least one intrinsic capacity domain at baseline, most commonly vision (77.7%), locomotion (53.1%), hearing (52.5%), and psychological (27.3%). Sedentary behavior was frequent (77%). Expert frailty scores correlated with the Fried phenotype, whereas associations with self-rated health were weaker. Models based on sensor parameters alone showed limited ability to reproduce Fried-defined frailty, while multimodal models integrating clinical and questionnaire variables improved discrimination. Over 6 months, kinesiotherapy and regular physical activity were associated with improved postural control metrics (including center-of-pressure features and mediolateral sway), while changes in gait speed were modest. Conclusions: An IoT-supported platform combining quantitative gait, balance, and grip strength measures with targeted questionnaires is feasible in outpatient primary care and yields frailty estimates broadly consistent with GP assessment. However, subjective and clinical inputs remain essential to capture psychological aspects of frailty not fully reflected by sensor signals alone. These findings support scalable frailty screening and longitudinal monitoring, and warrant validation in larger samples, including deployment by trained non-medical personnel and integration into precision-prevention pathways
Understanding the role of turbulence and biofilm on low density microplastic dynamics: An experimental approach towards natural conditions
No full textInternational audienc
Grands modèles de langue pour la détection de pathologies psychiatriques : promesses, réalité, et enjeux
No full textInternational audienceLes troubles de la santé mentale concernent une part croissante de la population. Face aux difficultés rencontrées par nos systèmes de santé, les technologies de TAL ont pu être présentées comme des alternatives prometteuses, en particulier pour l'aide au diagnostic. Pour comprendre la réalité des contributions apportées, nous avons effectué une revue de la littérature sur l'utilisation de LLM pour la détection de pathologies psychiatriques. Les résultats obtenus permettent d'analyser l'adéquation entre promesses mobilisées et état actuel de la recherche. Ceci nous amène finalement à évoquer les enjeux soulevés par l'intégration de LLM dans le domaine de la psychiatrie.</div
ARMAGNAC: A New Parametric Batch Normalization Layer for SPDNet Architecture
No full textSoumission EUSIPCO 2026This paper introduces ARMAGNAC (bAtch noR-Malisation pArametric Geometric harmoNic ArithmetiC), a new batch normalization layer for deep learning architectures on Symmetric Positive Definite (SPD) matrices. Unlike standard existing approaches that rely on computationally expensive Fréchet means with iterative eigenvalue decompositions, ARMAGNAC leverages a parametric mean that automatically adapts between simple arithmetic and harmonic means. This learnable parametrization enables the layer to adjust to the specific characteristics of the problem at hand while maintaining computational efficiency through explicit formulas of the backpropagation gradients. Experimental results demonstrate that ARMAGNAC correctly learn the parameter to use the most appropriate mean between arithmetic and harmonic on simulated data, and achieves superior performance results on real-data classification tasks
Cosmological constraints from a joint DESI DR1 Full-Shape and DR2 BAO
No full textInternational audienceWe present a cosmological analysis combining full-shape (FS) clustering measurements from the Dark Energy Spectroscopic Instrument (DESI) DR1 with baryon acoustic oscillation (BAO) measurements from DESI DR2. To achieve a robust combination that accounts for the correlation between the two data releases, we employ the ShapeFit compression method and estimate the joint covariance using EZmocks. This compressed approach inherently mitigates the prior volume effects that have previously dominated Bayesian constraints from DESI data with minimal external priors. Consequently, we obtain--for the first time within a Bayesian framework--reliable DESI-only constraints on extensions to CDM using only a Big Bang Nucleosynthesis prior on the baryon density and a wide prior on the spectral index. In flat CDM, we find , , and . For the CDM dynamical dark energy model, we measure and , improving constraints by relative to the analogous DR1 measurement and reducing the discrepancy with CDM to when compared to BAO only analyses. We also report competitive limits on the sum of neutrino masses and spatial curvature. This work demonstrates that the ShapeFit compression provides a prior-robust and computationally efficient pathway to constrain beyond-CDM physics with large-scale structure
Mapping NAT‐Containing Polar Stratospheric Clouds in the Antarctic Stratosphere Using Nadir Infrared Radiance Observations
No full textInternational audiencePolar stratospheric clouds (PSCs) play a key role in the formation of the Antarctic ozone hole. Here, we report the first observations of the most abundant PSC composition class, NAT-containing mixtures (one of whose components is nitric acid trihydrate, NAT, particles), from measurements of a passive nadir-viewing sounder, the Infrared Atmospheric Sounding Interferometer (IASI). We show the unambiguous presence of the characteristic peak signature of small NAT particles near 820 cm -1 in IASI spectra. By developing a detection method specific to the South Pole, we further demonstrate that NAT-containing PSCs can be detected systematically during the polar winter, despite a reduced sensitivity over the continent. This method allows us to obtain an innovative data set in terms of spatial and temporal sampling. With respect to the physicochemical processes at play in the Antarctic stratosphere, we show a high degree of consistency between the 2008-2024 time series of detected NAT-containing PSCs, stratospheric temperatures and total columns of nitric acid. Plain Language SummaryThe large ozone depletion occurring each year over the South Pole during the austral spring is strongly related to the presence of polar stratospheric clouds (PSCs). These clouds form at high altitudes under extremely cold conditions. In this work, we use satellite measurements from the Infrared Atmospheric Sounding Interferometer (IASI) to detect one common PSC composition class, NATcontaining mixtures. It is the first time that such detections are reported by an instrument of this kind, which offers a much denser spatial and temporal sampling than others, but does not provide vertical information on the presence of NAT particles. We obtain systematic observations during the polar winter, allowing us to study their spatial and temporal distributions. By analyzing their occurrence in relation to other atmospheric parameters (temperatures and nitric acid abundances), we show a high degree of consistency with processes governing the chemistry of the polar stratosphere during winter
Consistency–accuracy correlation in hard-prompted LLMs for entity and relation extraction: empirical findings from plant-health data
No full textInternational audienceAbstract As large language models (LLMs) become increasingly popular for information extraction (IE), concerns persist regarding the stability and reliability of their outputs. While accuracy has traditionally been the main evaluation metric, consistency—defined as the stability of model outputs across repeated runs—has recently been proposed as a complementary signal of reliability. In this work, we examine the relationship between accuracy and consistency in hard-prompted generative LLMs applied to entity and relation extraction. We conduct a systematic evaluation using four LLMs (GPT, DeepSeek, Qwen, Kimi) on the EPOP corpus, a plant-health dataset with rich entity types, long-range relations, overlapping relations, and strong argument constraints. To refine the interpretation of consistency, we distinguish between recoverable output variations—those that preserve the meaning of the extracted information—and critical ones that result in semantic errors. Our results show that while some positive correlation between accuracy and consistency exists, it is model-dependent and varies with task complexity. In structured prediction tasks, we show that consistency should be measured at the semantic level, ignoring superficial variations in format or wording. These insights have important implications for using self-consistency as a confidence filter and for designing reliable generative IE pipelines in specialized domains