Portail HAL UNIV-RENNES
Not a member yet
180610 research outputs found
Sort by
Late Breaking Results: SoC-FPGA HW Trojan leaking data through EM Covert Channel
No full textInternational audienceThis paper demonstrates an attack exploiting an Electromagnetic (EM) leak coming from the SoC-FPGA I/O. A covert channel is created by a dedicated Hardware Trojan controlling the EM emanations between the DDR3L SDRAM and the SoC-FPGA, exfiltrating sensitive data
Diapycnal Mixing and Tracer Dispersion in a Terrain‐Following Coordinate Model
No full textInternational audienceDiapycnal mixing, driven by small-scale turbulence, is crucial for the global ocean circulation, particularly for the upwelling of deep water masses. However, accurately representing diapycnal mixing in ocean models is challenging because numerical errors can introduce significant numerical mixing. In this study, we explore the diapycnal mixing in a high-resolution regional model of the North Atlantic subpolar gyre using the Coastal and Regional Ocean Community model (CROCO). CROCO uses terrain-following vertical coordinates that do not align with isopycnals. As such, tracer advection schemes produce spurious diapycnal mixing, which can nonetheless be reduced using rotated advection schemes. We focus on how different advection schemes and vertical resolutions affect numerical diapycnal mixing. Our approach includes online diagnostics of buoyancy fluxes and tracer release experiments to quantify the effective mixing, which combines parameterized and numerical diapycnal mixing. Our main results show that in flat-bottom regions, the effective diapycnal mixing is close to the parameterized mixing. However, in regions with steep topography, numerical mixing can locally significantly exceed parameterized mixing due to grid slope constraints imposed by the rotated mixing operator. While topography smoothing can mitigate this excessive mixing, it can also alter flowtopography interactions. In addition, while a higher vertical resolution reduces the numerical mixing induced by the vertical tracer advection, it can also increase numerical mixing in steep regions by introducing a stronger constraint on the grid slope. These results underscore that diapycnal mixing representation in a numerical model requires balancing high resolution and topographic smoothing with the control of numerical errors
From the Hazards of Death to the Hazards of Indication Creep for TAVR
No full textInternational audienc
Herds From Video: Learning a Microscopic Herd Model From Macroscopic Motion Data
No full textInternational audienceFigure 1: Our method can simulate individual agents to replicate herd behaviour learnt from a video containing many animals. [Left and middle] The original video (lower-right) and our simulation (upper-left), optimized to fit macroscopic density and velocity fields over a coarse grid. [Right] An authored simulation in which a herd transitions between the two illustrated behaviours, featuring narrow and broad formations.</div
Détection des Données Hors Distribution : Une Approche Basée sur un Auto-Encodeur Variationnel Structuré
No full textNational audienceLes modèles d’intelligence artificielle (IA) sont généralement entraînés sur des ensembles de données représentatifs de leur tâche. Cependant, leur performance peut être gravement affectée lorsqu’ils rencontrent des données hors distribution (OoD). Cet article explore une approche fondée sur un auto-encodeur variationnel structuré, permettant de mieux distinguer les données OoD des données légitimes sans nécessiter un ensemble d’entraînement OoD spécifique mais en tirant partie des échantillons OoD disponibles lors du déploiement. En particulier, nous montrons l’utilité d’un travail en deux étapes, et d’un "padding" à l’aide de données OoD connues et de types divers, ce qui permet une amélioration sensible des performances
Quelle diffusion dans le partenariat UE-Vietnam et UE-ASEAN de "la nouvelle approche pour une économie bleue durable dans l’Union européenne ? "
No full textInternational audienc
Chimpanzee mothers, but not fathers, influence offspring vocal–visual communicative behavior
No full textInternational audienceFace-to-face communication in humans typically consists of a combination of vocal utterances and body language. Similarly, our closest living relatives, chimpanzees, produce multiple vocal signals alongside a wide array of manual gestures, body postures and facial expressions. In humans, the ontogenetic development of communicative behavior is known to be heavily influenced by the child’s primary caretakers. In chimpanzees, the extent to which communicative behavior is learned, as opposed to genetically inherited, remains openly debated. Here, we address this issue within the context of multi-modal communication by investigating kinship patterns in the production of visual behaviors alongside vocal signals in wild chimpanzees from the Kanyawara community, Uganda. We report a similarity in the number of visual behaviors combined with vocal signals between individuals who are related via their mother, while no similarity is observed between paternal relatives, in line with the observation that chimpanzee mothers constitute the primary caretakers, while fathers are not involved in parenting. We conclude that the development of this aspect of multi-modal communicative behavior is unlikely to be genetically driven and is rather a result of learning via exposure to social templates, akin to processes involved in the acquisition of human communication
Integrating and Analysing Occupational Health Data Using a Multi-Ontology Approach
No full textInternational audienceA variety of occupational data are collected by health organisations to investigate workplace exposures encountered by workers in their occupational activities and the potential health effects that may arise. These datasets have diverse characteristics and are not inherently designed to interoperate. However, they contain complementary information, which, when analysed collectively, can provide a broader perspective on high-risk occupational scenarios and inform targeted prevention strategies. The objective of this study is to develop a methodology to integrate and analyse heterogeneous French data. For this, ten French occupational databases, provided by six French institutes were used. An Ontology-Based Data Integration approach was employed, involving the mapping of data sources to a domain-specific ontology, namely the Adapted Occupational Exposure Ontology. Four additional ontologies were utilised: the Occupational Exposure Thesaurus, which categorises occupational exposures and hazards; the International Classification of Diseases, which classifies health disorders and diseases; the French Nomenclature of Activities, which identifies activity sectors in France; and the Professions and Socio-professional Categories, which defines occupational classifications. Data integration is primarily achieved through the concept of the “occupational group”, defined as a group of individuals sharing the same sex, occupation, and activity sector. Two case studies derived from the integrated dataset are presented: (1) a quantitative analysis identifying occupational groups at highest risk and most affected by diseases; and (2) a qualitative analysis evaluating the consistency of exposure and disease-related information. The construction sector was selected for these case studies due to its significance in occupational health research and the availability of substantial, relevant data. This methodological approach structures all the data and enables various analysis methods to be designed and implemented, making it possible to envisage targeted responses to current and emerging occupational health problems using specialised tools and queries
Évaluation des effets de la sécheresse et de l'urbanisation sur les arbres urbains à l'aide de la série temporelle Sentinel-2
No full textInternational audienceUrban trees provide essential ecosystem services, including temperature regulation, carbon storage, and biodiversity conservation, which are crucial for enhancing urban living conditions. However, they are exposed to various stress factors in urban environments, such as limited light, restricted growth space, and increased exposure to pollutants. Moreover, climate forecasts predict an increase in the frequency and severity of droughts and drought events which can lead to tree defoliation, increased sensitivity to pests and pathogens and therefore increase tree mortality. Quantifying and monitoring the response of urban trees to stress factors can be used as a proxy for tree health assessment.The effects of stress factors (drought, urban conditions) on urban trees dynamics have already been observed in situ with dendrochronology and biochemical analysis. These effects have been shown on a large scale using remote sensing data, but the coarse spatial resolution of the data used (MODIS, 250m) make it impossible to discriminate species that suffer from stress from those that are more resilient. We first investigated tree dynamics in Rennes (France) at fine spatial scale using Sentinel-2 time series to analyze the impact of drought events and urbanization on tree health. Specifically, we analyzed five deciduous tree species (Platanus Acerifolia, Acer Platanoides, Fraxinus Excelsior, Quercus Rubra and Quercus Robur) during two years characterized by very contrasting climatic conditions (2021 and 2022) and along an urban-rural gradient. Vegetation dynamics were monitored using a vegetation index (ARVI). Phenological, productivity and disturbance metrics which characterize respectively key dates and growth-cycle duration, tree growth and productivity, and intra-annual anomalies in tree-dynamics, were derived from the ARVI filtered time-series. Drought events were characterized using a standardized precipitation index derived from climate data and urban intensity was determined at pixel-scale using the Copernicus’ Imperviousness Density data. Temporal metrics values were aggregated at patch-scale (group of contiguous trees of the same species). The results showed a notable extension in the length of the growing season and maturity periods for four species in 2022 (the driest year), driven primarily by delayed senescence and end-of-season dates. Productivity metrics displayed mixed responses, with some species showing reduced growth under drought, while others, like Platanus Acerifolia, exhibited increased productivity, suggesting potential resilience mechanisms. Elevated disturbance levels were observed in 2022, indicating higher stress conditions compared to the previous year. Furthermore, the results outlined that urban intensity was most often correlated with extended growing seasons and altered productivity dynamics. Based on these results, we then developed an exploratory approach at the city scale to identify trees with unusual temporal profiles that correspond to exacerbated trees responses to stress factors. We determined a reference profile for each tree species, corresponding to the average profile of trees growing in similar and contrasting urban conditions, and compared tree profiles to this reference profile based on statistical analysis. This analysis was conducted for period 2016-2024 to monitor tree disturbance levels. This tool can be used by urban tree managers to locate, guide and schedule in situ investigations to monitor tree health and potentially identify trees in decline. This study underscores the value of Sentinel-2 time series in supporting urban tree management and policy decisions amidst changing environmental and climate conditions
Reliability of earthworm data from citizen science: Lessons from 7 years of a French national monitoring protocol
No full textInternational audienceMonitoring biodiversity is seldom comprehensive, as the spatio-temporal resolution needed to accurately reflect dynamic changes of these communities in diverse environments is often lacking. Citizen science offers a promising tool to help fill these gaps, engaging a wider audience in monitoring efforts and thus enhancing our understanding of earthworm ecology. However, a significant challenge arises as earthworms are difficult to identify to the species level in the field by non-experts, necessitating the use of morphotypes as taxonomic proxies. This study evaluates the reliability of earthworm classification into four earthworm morphotypes within the '500 ENI' (Non-intended Effects) Monitoring Network in France. The network relies on annual sampling conducted in agricultural lands by non-specialist participants with subsequent identification verification by earthworm taxonomists. Analyzing >48,000 individual earthworms collected over 950 plots, we calculated two indices: the misclassification rate (MR) and the undetected rate (UR) to assess the reliability of classification into earthworm morphotypes. The results indicated an average MR of 28 % and an average UR of 32 %, which both varied according to morphotypes. Endogeics had lower error rates compared to epigeics, anecics with a red anterior, and anecics with a black anterior. Our findings underscored the significant impact of sampler experience and earthworm community composition on the reliability of classification of individuals into morphotypes by citizens. The results highlight the critical need for enhanced support and guidance for participants with limited experience. Furthermore, we recommend providing additional training or resources to aid in morphotype classification, especially for earthworm communities exhibiting low abundance, low adult proportion, or low morphotype diversity. Encouraging participants to sample during periods favorable for detecting reliable total and adult abundances would also help optimize morphotype detectio