Archive ouverte de Centrale Lyon
Not a member yet
32420 research outputs found
Sort by
Characteristics analysis of silicone gel electrical treeing for power device packaging under high temperature coupled square wave pulse electric field
No full textInternational audienc
Apprentissage automatique avec des prédicteurs environnementaux pour prévoir les visites et les admissions à l'hôpital : une revue systématique.
No full textInternational audienceMany studies have demonstrated a correlation between environmental monitoring data and healthcare service demand, highlighting its contribution to the global problem of emergency department crowding. To address this problem, forecasting models are essential for resource allocation and general management to improve patient outcomes. Machine Learning (ML), especially Deep Learning (DL), offers promise for forecasting patient volume. In this work, we present a systematic review of the use of ML to predict the health impacts of environmental exposures in the context of hospital visits and admissions. Standardized tools for conducting a systematic review were used, including the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS), and Prediction model Risk Of Bias ASsessment Tool (PROBAST). The search included studies from 2012 to 2025. We focus on answering how ML has been applied and what the major environmental predictors are used. As a result, 36 studies were retained from PubMed, Embase, and IEEE Xplore databases. We found that many studies exhibited a high risk of bias due to poor handling of missing values, inadequate outcome definitions, biased participant selection, and a low number of events per variable. Additionally, we found that the most used air pollutants and meteorological variables were PM2.5, PM10, NO2, SO2, CO, O3, and temperature. Furthermore, the most common models were Random Forest and feed-forward neural networks. In addition, land use, remote sensing, demographic, and socioeconomic data offer promising avenues for improving model performance.De nombreuses études ont démontré une corrélation entre les données de surveillance environnementale et la demande de services de santé, mettant en évidence leur contribution au problème mondial de la surcharge des services d’urgence. Pour résoudre ce problème, les modèles de prévision sont essentiels pour l’allocation des ressources et la gestion générale afin d’améliorer les résultats pour les patients. L’apprentissage automatique (ML), en particulier l’apprentissage profond (DL), offre des perspectives prometteuses pour la prévision du volume de patients. Dans ce travail, nous présentons une revue systématique de l’utilisation du ML pour prédire les impacts sanitaires des expositions environnementales dans le contexte des visites et admissions hospitalières. Des outils standardisés pour la conduite d’une revue systématique ont été utilisés, notamment les Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), la Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS), et le Prediction model Risk Of Bias ASsessment Tool (PROBAST). La recherche a inclus des études allant de 2012 à 2025. Nous nous concentrons sur la manière dont le ML a été appliqué et sur les principaux prédicteurs environnementaux utilisés. En conséquence, 36 études ont été retenues à partir des bases de données PubMed, Embase et IEEE Xplore. Nous avons constaté que de nombreuses études présentaient un risque élevé de biais en raison d’une mauvaise gestion des valeurs manquantes, de définitions inadéquates des résultats, d’une sélection biaisée des participants et d’un faible nombre d’événements par variable. De plus, nous avons constaté que les polluants atmosphériques et variables météorologiques les plus utilisés étaient le PM2.5, le PM10, le NO2, le SO2, le CO, l’O3 et la température. En outre, les modèles les plus courants étaient la forêt aléatoire (Random Forest) et les réseaux neuronaux à propagation avant. De plus, l’utilisation du sol, la télédétection, ainsi que les données démographiques et socioéconomiques offrent des perspectives prometteuses pour améliorer les performances des modèles
Algèbres enveloppantes d’algèbres de Lie graduées simples qui sont noethériennes
No full textIt is shown that if the universal enveloping algebra of a simple Z n -graded Lie algebra is Noetherian, then the Lie algebra is finitedimensional.Nous montrons que lorsque une algèbre de Lie simple Zn-graduée a une algèbre enveloppante noethérienne, elle est de dimension finie
Analysis of local current density, temperature, and mechanical pressure distributions in an operating PEMFC under variable compression
No full textInternational audienceThis article investigates the impact of mechanical compression on local phenomena within an operating PEMFC with a large active area (225 cm2). It explores the distributions of current density, temperature, and mechanical pressure, building on previous global characterisations (cell voltage, polarisation curves, and EIS). The study finds that mechanical compression (0.35–1.55 MPa) enhances the uniformity of current density and temperature distribution, reducing the risk of hotspots that can impair PEMFC performance and durability. The spatial analysis reveals that this homogenisation effect is mainly due to improved pressure distribution with increased compression. The regions with higher mechanical pressure correlate with higher local current density and temperature, which improves performance by reducing ohmic resistance. However, excessive compression at high current density and relative humidity can lead to water management issues. Overall, the results support the positive effect of pressure homogenisation on PEMFC performance, as observed in previous studies
Player-Centric Shot Maps in Table Tennis
No full textInternational audienceShot maps are popular in many sports as they typically plot events and player positions in the way they are collected, using a pitch or a table as an absolute coordinate system. We introduce a variation of a table tennis shot map that shifts the point of view from the table to the player. This results in a new reference system to plot incoming balls relative to the player's position rather than on the table. This approach aligns with how table tennis tactical analysis is conducted, focusing on identifying empty spaces and weak spots around the players. We describe the motivation behind this work, built through close collaboration with two table tennis experts, and demonstrate how this approach aligns with the way they analyze games to reveal key tactical aspects. We also present the design rationale and the computer vision pipeline used to accurately collect data, leveraging recent table tennis data extracted from broadcast videos. Our findings show the technique enables to capture insights that were not visible with the absolute coordinate system
Vers une Extraction Automatique de Structures Spatiales Statiques pour le Français: Application au corpus parallèle EN80jours
No full textInternational audienceBasic Locative Structures (BLCs) have long served as a controlled framework for investigating space in language and cognition. Recently, (Viechnicki et al., 2024) proposed an automatic BLC extraction method from parallel corpora using English as the pivot language. In this study, we adapt and apply this methodology to the EN80Jours parallel corpus, with French serving as the pivot language. By combining syntactic and lexical pattern searches with few-shot learning using Pretrained Language Models to filter for spatial expressions, we achieve precise retrieval of a refined set of locative constructions. Finally, we make initial observations of the inter-linguistic variability of spatial markers within BLCs by aligning French spatial markers with their German and English counterparts. This preliminary analysis highlights challenges in aligning BLCs across languages and suggests the need for further exploration of linguistic variations and the integration of lexical rules to refine classification
GRANITE : a Byzantine-Resilient Dynamic Gossip Learning Framework
No full textGossip Learning (GL) is a decentralized learning paradigm where users iteratively exchange and aggregate models with a small set of neighboring peers. Recent GL approaches rely on dynamic communication graphs built and maintained using Random Peer Sampling (RPS) protocols. Thanks to graph dynamics, GL can achieve fast convergence even over extremely sparse topologies. However, the robustness of GL over dy- namic graphs to Byzantine (model poisoning) attacks remains unaddressed especially when Byzantine nodes attack the RPS protocol to scale up model poisoning. We address this issue by introducing GRANITE, a framework for robust learning over sparse, dynamic graphs in the presence of a fraction of Byzantine nodes. GRANITE relies on two key components (i) a History-aware Byzantine-resilient Peer Sampling protocol (HaPS), which tracks previously encountered identifiers to reduce adversarial influence over time, and (ii) an Adaptive Probabilistic Threshold (APT), which leverages an estimate of Byzantine presence to set aggregation thresholds with formal guarantees. Empirical results confirm that GRANITE maintains convergence with up to 30% Byzantine nodes, improves learning speed via adaptive filtering of poisoned models and obtains these results in up to 9 times sparser graphs than dictated by current theory
Antiferroelectric PUND measurements on ZrO2
No full textInternational audienceThis work introduces the AFE-PUND measurement method, an originalpositive up negative down (PUND) protocol intended to accurately study antiferroelectric(AFE) thin-film behavior. As for its FE counterpart, the AFE-PUND method isolatesswitching currents from nonswitching contributions, enabling a precise extraction ofsaturation polarization and coercive fields from hysteresis loop curves. In this paper, AFEPUND was deployed on AFE ZrO2 films of varying thicknesses. The results reveal thatsaturation polarization increases with film thickness, indicating enhanced domain stability,while endurance cycles showcase the wake-up effect and its eventual fatigue-induceddegradation in thicker films. Similarly, coercive fields decrease with thickness, reflectingreduced switching barriers and a sharper transition between the tetragonal andorthorhombic phases. AFE-PUND establishes itself as an extremely valuable method foradvancing the understanding and optimization of AFE materials
Sorting Circulating Tumor Cells: A Low Flow Microfluidic Pre-Enrichment Function for Improved Separation in Serial Two-Stage Sorting Device
No full textInternational audienceThe isolation of Circulating Tumor Cells (CTCs) directly from blood by liquid biopsy could lead to a paradigm shift in clinical cancer care by enabling earlier diagnosis, more accurate prognosis and personalized treatment. Nevertheless, the specific challenges of CTCs, including their rarity and heterogeneity, have so far limited the use of CTCs in clinical studies. Currently, no device fully meets the requirements of high recovery, high purity, short processing time and ease of use for end-users. A promising new strategy involves combining a higher throughput but less specific pre-enrichment step based on size sorting together with a highly specific but slower immunomagnetic sorting. This approach requires the initial function to operate at lower flow rates than commonly used to connect the two functions in series. In this context, we developed a Dean spiral microfluidic device, optimized for sorting 10µm and 15µm beads by size. We showed that it successfully separates mimicking CTCs from white blood cells at low flow rates (<100 mL/h). a</div
Optimal operational planning of biomass district heating: Adaptation to air pollution episodes with LCA-based dynamic penalization
No full textInternational audiencehe pervasive impacts of climate change are reshaping both the development of new technologies and the adaptation of existing ones, especially for energy systems. Besides the reduction of their environmental footprint, they must cope with extreme weather events. However, for this large-scale system, the adaptation to new climate events is subject to several operational constraints and economic challenges. In this work, the adaptation of a common configuration of French district heating networks is evaluated. The energy portfolio consists of biomass, waste incineration, gas and thermal energy storage. We examine how the system can be adapted to face pollution peaks of particulate matter during anticyclonic events that occur during the heating season in France. A novel dynamic penalization for pollutant emissions from biomass boilers is derived from its variable load operation. It is then used in a multicriteria planning optimization. This approach provides a new perspective to manage emissions during air pollution peaks, an area that has not been thoroughly explored in district heating adaptation strategies. The results show that the network is able to avoid using biomass at low loads during pollution episodes against a slight increase of 2.1% in the operational costs. In addition to performing life cycle impact assessment, the relevance of our approach is tested by comparing different penalization formulations and by varying the thermal energy storage capacity