HAL Evry
Not a member yet
19237 research outputs found
Sort by
An efficient column-and-constraint generation algorithm for solving the adaptive robust elective surgery problem under uncertainties in surgery duration, length of stay, and emergency arrivals
No full textInternational audienceThis paper addresses the robust elective surgery problem in the context of Operating Room (OR) planning, incorporating downstream resource constraints related to the Intensive Care Unit (ICU) and uncertainties in surgery duration, length of stay in the ICU, and emergency arrivals. We propose a novel two-stage robust optimization approach, using the “here-and-now” and “wait-and-see” decision-making principles. In the first stage, decisions on the master surgical schedule and surgical case assignment problem are made, allocating patients, surgeons, and specialties to OR sessions under the block scheduling strategy, using symmetry breaking inequalities. The second stage addresses multiple uncertainties and aims to minimize the costs in the worst-case scenarios, considering session overtime, emergency costs, and the costs of denied ICU beds. Polyhedral uncertainty sets and structural properties enable the use of the Column-and-Constraint Generation algorithm for efficient problem resolution. Computational experiments using real data from a medium-sized French hospital demonstrate the approach’s superior resource utilization and computational efficiency over the cutting-plane method. Value at Risk, Conditional Value at Risk, and Monte Carlo simulation are used to assess robustness, aiding in parameter adjustments for risk prevention and cancellations. This research contributes to enhancing decision-making in elective surgery planning under multiple uncertainties, offering practical insights for improving hospital productivity
NESO Based Ultra-Local Model Predictive Control for Autonomous Vehicle Path Tracking and Roll Stability
No full textInternational audienceThis paper presents a model predictive controller for autonomous vehicles based on ultra-local models and a Nonlinear Extended State Observer (NESO) that simultaneously address path tracking performance and roll stability. The controller consists of several key components: first, ultra-local models are developed to represent the vehicle lateral error dynamics and roll dynamics. Next, a nonlinear extended state observer is designed to estimate integrated disturbances and unmodeled states within the ultra-local models. The parameters of the observer are optimized using the Butterfly Optimization Algorithm (BOA) in a simulation environment to enhance estimation accuracy. Finally, a model predictive control (MPC) scheme is implemented to minimize both lateral tracking error and roll angle, effectively addressing both path tracking performance and roll stability. The effectiveness of the proposed controller is validated through comprehensive comparative simulations conducted on a MATLAB/Simulink and Carsim co-simulation platform
Entry and exit of Fragilariopsis cylindrus diatoms from the polar night: evidence for chromatin-mediated genome control
No full textRegulation of the epigenome landscape plays a crucial role in adaptation to environmental changes. However, the molecular mechanisms by which epigenome landscapes are established and tuned to adjust the cellular status to environmental cues remain poorly understood, especially in unicellular photosynthetic eukaryotes. Polar microalgae must adapt to extreme variations in light and temperature for survival and therefore represent excellent models to understand the diversity of gene regulatory mechanisms, yet chromatin-based mechanisms have never been explored in these organisms. In this study, we combined genome-wide profiling of histone and DNA modifications with transcriptomic analyses of the marine polar diatom Fragilariopsis cylindrus upon entry and exit of a 3-month-long polar night. We show that, in agreement with its established role in transcription facilitation, histone H2B monobiquitination (H2Bub) is modulated at active genes upon prolonged darkness to light re-exposure. In contrast, histone H3 lysine 27 trimethylation (H3K27me3) and DNA cytosine methylation (CG methylation), two modifications typically associated with the silencing of genes or transposable elements (TEs), remained stable throughout the transitions, reinforcing their role in transcriptional repression, particularly at TEs but also at a few genes. We further demonstrate that H3K27me3 and DNA methylation are physically associated at TEs, probably reflecting a dual locking system enabling F.cylindrus cells to cope with the invasive nature of these genetic elements under extreme environmental conditions. These findings enhance our understanding of genome and epigenome regulation in diatoms in response to environmental variability and open new avenues for exploring the role played by chromatin-level regulation in the unique evolutionary history of polar unicellular eukaryotes
MIXSIM3D : Une Nouvelle Méthode d'Apprentissage Curriculum Contrastif 3D Appliquée à la Physique des Roches
No full textInternational audienceIn this paper, we introduce a new method named MixSim3d, a deep learning self-supervised approach designed tolearn meaningful representations and extract relevant parameters from 3D images. It combines curriculum learning and contrastivelearning (CL) to improve the robustness of the learned embedded representation of input 3D images. Here, we apply this methodin the context of Digital Rocks Physics (DRP) to predict properties, such as porosity and permeability, from the observed 3Ddatasets. Our evaluation shows that the proposed Mixsim3d method can obtain promising results, outperforming existing baselineself-supervised approaches in particular scenarios.Dans cet article, nous introduisons une nouvelle méthode appelée MixSim3d, une approche d'apprentissage profond auto-supervisée conçue pour apprendre des représentations latentes pour des tâches de régression, à partir d'images 3D. Elle combine l'apprentissage curriculum (CL) et l'apprentissage contrastif pour améliorer la robustesse de la représentation latente apprise sur les images 3D en entrée. Ici, nous appliquons cette méthode dans le cadre de la Physique des Roches Numérique (PRN) pour prédire des propriétés telles que la porosité et la perméabilité à partir des ensembles de données 3D observées. Notre évaluation montre que la méthode MixSim3d proposée peut obtenir des résultats prometteurs, surpassant certaines approches auto-supervisées de référence existantes dans des scénarios particuliers
Dual Focus Multiscale Attention for Object Detection in Mixed Reality: Leveraging Customizable Synthetic Datasets
No full textInternational audienceWe propose a novel object detection framework tailored for mixed reality (MR), combining a customizable synthetic dataset with a lightweight attention-enhanced detection model. Our dataset generation pipeline synthesizes planetary and telescope foregrounds with hybrid real-synthetic backgrounds, enabling robust learning across variable lighting and occlusion scenarios—challenges com- mon in educational MR environments. At the core of our architecture is the Dual Focus Multiscale Attention (DFMA) module, which simultaneously refines spatial and channel-wise features at multiple scales. Integrated into a YOLO-based (You Only Look Once) backbone and FPN, DFMA significantly improves feature discrimination while preserving real-time efficiency. On MS COCO our model improves mean Average Precision (mAP) across Intersection over Union (IoU) thresholds from 0.5 to 0.95 ([email protected]:0.95) over state-of-the-art nano detectors from 39.3% to 41.3% (± 2%) at only +6% params and +3% GFLOPs, with a notable reduction in false positives on visually similar, low-textured objects. We further demonstrate real-time deployment in a Unity-based MR application, highlighting the system’s effectiveness in immersive astronomy-focused educational scenarios. Our results underscore the potential of synthetic data and multiscale attention to bridge accuracy, speed, and realism in next generation MR systems
Embedding digital responsibility: an exploratory analysis of AI policies in higher education
No full textInternational audienceWith AI use rising in academic environments, AI governance has become a critical discussion. This paper aims to understand the extent to which the top 30 ranked universities embed digital responsibility principles within their AI governance frameworks. To ensure consistent and transparent analysis, only universities with publicly accessible AI governance documents were included. This criterion narrowed the sample from 30 to 21 institutions. Guided by Trier's (2022) proposed framework of eight principles of digital responsibility: Sustainability, Participation, Functionality, Data Privacy, Transparency, Fairness, Norms/Values, and Accountability, we conducted a thematic analysis of the narrowed set of AI policy documents from top 21 ranked universities to examine the presence and depth of these principles within universities’ AI governance frameworks. The preliminary findings reveal varying degrees of commitment to digital responsibility, with some principles consistently emphasised, such as Data Privacy and Accountability, while others such as Functionality and Participation are frequently overlooked. These inconsistencies provide insights into how leading HEIs approach responsible AI governance and highlight areas for strengthening responsible AI governance. Improved AI governance is essential to ensuring a safe and ethical environment for both students and educators, mitigating the risks of AI usage, maximising the benefits of AI, and providing transparency to institutional expectations regarding usage of AI through the HEI ecosystem. This study contributes to growing discourse on responsible AI and AI governance, and provides insights into best practices and recommendations for the integration of AI into education
Un fichier photographique de prostitué·es dans un commissariat de police : contrôle mobile et subversion du modèle (1973-1983)
No full textInternational audienc
Polyploidization leads to salt stress resilience via ethylene signaling in citrus plants
No full textData availability: All high-throughput sequencing data generated in this study have been deposited in the SRA database under the accession no.: PRJNA1159448, which includes the WGS-seq data, RNA-seq data, the BS-seq data and the ATAC-seq dataInternational audiencePolyploidization is a common occurrence in the evolutionary history of flowering plants, significantly contributing to their adaptability and diversity. However, the molecular mechanisms behind these adaptive advantages are not well understood. Through comprehensive phenotyping of diploid and tetraploid clones from Citrus and Poncirus genera, we discovered that genome doubling significantly enhances salt stress resilience. Epigenetic and transcriptomic analyses revealed that increased ethylene production in the roots of tetraploid plants was associated with hypomethylation and enhanced chromatin accessibility of the ACO1 gene. This increased ethylene production activates the transcription of reactive oxygen species scavenging genes and stress‐related hormone biosynthesis genes. Consequently, tetraploid plants exhibited superior root functionality under salt stress, maintaining improved cytosolic K + /Na + homeostasis.To genetically validate the link between salt stress resilience and ACO1 expression, we generated overexpression and knockout lines, confirming the central role of ACO1 expression regulation following genome doubling in salt stress resilience.Our work elucidates the molecular mechanisms underlying the role of genome doubling in stress resilience. We also highlight the importance of chromatin dynamics in fine‐tuning ethylene gene expression and activating salt stress resilience pathways, offering valuable insights into plant adaptation and crop genome evolution
The Demographic History of Populations and Genomic Imprinting have Shaped the Transposon Patterns in Arabidopsis lyrata
No full textData availability: Nanopore long-read sequencing data for the three plant individuals—RON5.9 (sample SAMN43781236), TSS14.18 (sample SAMN43781228), and INDP15.1 (sample SAMN43781226)—are available in EBIN/ENA at https://www.ebi.ac.uk/ena, and can be accessed under project PRJEB80457.International audienceAbstract Purifying selection is expected to prevent the accumulation of transposable elements (TEs) within their host, especially when located in and around genes and if affected by epigenetic silencing. However, positive selection may favor the spread of TEs, causing genomic imprinting under parental conflict, as genomic imprinting allows parent-specific influence over resource accumulation to the progeny. Concomitantly, the number and frequency of TE insertions in natural populations are conditioned by demographic events. In this study, we aimed to test how demography and selective forces interact to affect the accumulation of TEs around genes, depending on their epigenetic silencing, with a particular focus on imprinted genes. To this aim, we compared the frequency and distribution of TEs in Arabidopsis lyrata from Europe and North America. Generally, we found that TE insertions showed a lower frequency when they were inserted in or near genes, especially TEs targeted by epigenetic silencing, suggesting purifying selection at work. We also found that many TEs were lost or got fixed in North American populations during the colonization and the postglacial range expansion from refugia of the species in North America, as well as during the transition to selfing, suggesting a potential “TE load.” Finally, we found that silenced TEs increased in frequency and even tended to reach fixation when they were linked to imprinted genes. We conclude that in A. lyrata, genomic imprinting has spread in natural populations through demographic events and positive selection acting on silenced TEs, potentially under a parental conflict scenario
