Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
First long‐term trajectory of an ocean sunfish (Mola mola L .) from the northwestern Mediterranean
No full textInternational audienceThe ocean sunfish is a large fish for which many aspects of its ecology and biology are still poorly known. Electronic tagging was used to provide the first information on the movements of an ocean sunfish from the northwestern Mediterranean. The sunfish moved towards the Gibraltar strait over the year and displayed substantial movements in the water column. The potential of the tagging technique employed for studying its behavior and long‐term migratory dynamics, and assessing the post‐release survival of ocean sunfish is highlighted
Anomaly Detection in Drone Videos for Preventive Maintenance of Power Lines
No full textInternational audiencePower lines are composed of various components that can deteriorate over time due to use. To detect potential issues with these components and prevent costly network outages, aerial drones are increasingly being utilized. They allow for the rapid inspection of large distances and provide a clear view of the different components and their defects. However, the manual analysis of flight videos by experts is labor-intensive. Moreover, the wide variety of anomalies that can cause network interruptions makes it impractical to develop dedicated automated solutions for each type, particularly due to the limited number of examples available for many anomalies. We thus propose a method to automatically detect anomalies in scenarios where no prior information about the visual appearance of anomalies is available, using drone-acquired videos. The approach relies on the computation of an anomaly score based on a generic feature vector. Results demonstrate that this approach is effective in scenarios without any examples of anomalies and requires very limited computational resources for learning
Background-invariant re-identification of dogs from camera-trap videos in non-controlled environments
No full textSource Agritrop Cirad (https://agritrop.cirad.fr/616519/)International audienceThis paper addresses the general problem of re-identification in natural conditions with multiple camera traps, poor video quality and small datasets. We focus on generalizable re-identification of dogs in cross-camera setups, adapting from short-term to long-term scenarios. Long-term re-identification across multiple cameras presents challenges due to variations in background, camera angles, and lighting conditions. While realistic, few animal re-identification methods are tested under such settings, mainly due to the lack of datasets and high complexity of annotation. Short-term datasets are often used to train re-identification networks, since they can be simply generated through web scraping algorithms. We introduce two publicly available datasets: the YT-BB-Dog, a short-term dataset with 2723 dogs from YouTube videos, and the Sibetan, a long-term dataset featuring 59 dogs recorded over 5 days and 12 cameras placed in Sibetan, Bali, Indonesia. Our goal is to use the YT-BB-Dog to train a feature extractor robust to covariate shifts, enabling better generalization in unknown domains. Our experiments revealed that state-of-the-art (SOTA) methods trained on the YT-BB-Dog are heavily influenced by background variations and perform poorly on complex scenarios like Sibetan. To address this, we propose Background Invariant Feature extractOR (BIFOR), a three-step method that leverages a novel mini-batch sampling technique with triplet loss and online hard mining. BIFOR achieves SOTA performance on Sibetan, improving rank-1 accuracy of the baseline by more than 9%. We also present a complete pipeline combining detection, tracking, and re-identification based on BIFOR
Carbon-Neutral Computing at the Edge: Scalable Greedy—LP Optimization of Green Data Centers
No full textInternational audienceAs energy demands in cloud computing surge, efficient task scheduling and carbon-free energy use in mini data centers is essential. Solving scheduling problems with mixedinteger linear programming optimization methods is computationally intensive and less scalable. This paper proposes a two-step approach combining a greedy algorithm with linear programming to optimize virtual machines scheduling and energy management. We show via simulations that our method preserves solution quality, while accelerating the calculation process by 99.6%. While the literature method requires 6 hours to solve a reference system design, our solution handles 18 times larger designs in 20 minutes only, demonstrating its scalability
Analyse des performances des méthodes d’apprentissage de représentations pour l’alignement d’entités : jeux de données de référence versus données du monde réel
No full textInternational audienceRepresentation learning for entity alignment (EA) aims to identify entities in different knowledge graphs (KGs) that refer to the same real-world object by comparing their embedding similarity. Although many EA models perform well on synthetic benchmark datasets, this performance does not always transfer to real-world, incomplete, and domain-specific data. A systematic comparison between synthetic benchmarks and original heterogeneous datasets is still limited. Many EA models also restrict the alignment search space to validation entities, limiting coverage of real KG content. Within this setting, our results show that embedding-based EA models continue to face generalization challenges in realistic large-scale KG search spaces. We evaluate several competitive EA models-commonly tested on benchmarks such as DBP15K-on multiple real-world heterogeneous datasets. The experiments reveal a performance decrease when moving beyond synthetic benchmarks, indicating that current models do not fully capture the characteristics of real data. We also analyze semantic similarity and profiling features of the datasets to help explain these differences. This study outlines practical limitations of embedding-based EA methods and provides insights for developing approaches that better handle the variability and complexity found in real-world KG alignment tasks.L’apprentissage de représentations pour l’alignement d’entités (EA) vise à identifier des entités dans différents graphes de connaissances (KG) qui renvoient au même objet du monde réel en comparant la similarité de leurs représentations vectorielles. Bien que de nombreux modèles d’EA obtiennent de bonnes performances sur des jeux de données de référence synthétiques, ces performances ne se transfèrent pas toujours aux données du monde réel, incomplètes et spécifiques à un domaine. Une comparaison systématique entre les benchmarks synthétiques et les jeux de données hétérogènes originaux reste encore limitée. De nombreux modèles d’EA restreignent également l’espace de recherche de l’alignement aux entités de validation, ce qui limite la couverture du contenu réel des KG. Dans ce contexte, nos résultats montrent que les modèles d’EA basés sur des représentations vectorielles continuent de faire face à des défis de généralisation dans des espaces de recherche de KG réalistes et à grande échelle. Nous évaluons plusieurs modèles d’EA compétitifs - couramment testés sur des benchmarks tels que DBP15K - sur plusieurs jeux de données hétérogènes du monde réel. Les expériences révèlent une baisse des performances lorsque l’on dépasse les benchmarks synthétiques, indiquant que les modèles actuels ne capturent pas pleinement les caractéristiques des données réelles. Nous analysons également la similarité sémantique et les caractéristiques de profilage des jeux de données afin d’aider à expliquer ces différences. Cette étude met en évidence les limites pratiques des méthodes d’EA basées sur des représentations vectorielles et fournit des pistes pour le développement d’approches capables de mieux gérer la variabilité et la complexité propres aux tâches d’alignement de KG du monde réel
Deep-Learning-Based Prediction of Occlusal Stresses on Teeth
No full textWe propose a deep-learning approach involving the training of two neural networks to enable real-time simulations while maintaining sufficient accuracy for educational purposes. The training data was generated from finite element simulations by varying parameters such as the location and magnitude of the force applied to the tooth.• We evaluated the performance of two neural networks. The first network was trained to predict the deformations of tooth elements after force application, from which the resulting stresses were then calculated. The second network directly predicted the stresses. Ultimately, the second network proved to be more accurate in stress prediction.• The approach focuses on visualizing the stresses experienced by teeth during mastication. A deep-learning method trained with biomechanical simulation data permits a real-time visualization. This proof of concept demonstrates the potential for developing educational tools to enhance understanding of occlusion through simulation results.</div
Justified Preference Aggregation in Agri-Food Systems: An Approach, Argumentation Methods, and Tools
No full textInternational audienceIn recent years, there has been a growing recognition of the need to ensure sustainability of agri-food systems, covering a variety of stakeholders and activities from production to waste management. Multi-Criteria Decision Assessment (MCDA) have emerged as crucial tools for sustainability assessment, but many do not consider stakeholder perspectives, which prevents their adoption and use. Our projects, NoAW (No Agricultural Waste) and AgriLoop (High-value products from agricultural residues through sustainable chains), address this issue by focusing on innovative valorization routes for agricultural waste by assessing stakeholder impact categories through participatory decision-making. This article proposes a novel methodology integrating computational social choice and argumentation techniques for achieving justified collective decision-making in agri-food systems. This methodology includes a theoretical framework and related tools, which facilitate the identification, analysis, and aggregation of preferences based on justifications
ClimBurst: A Dynamic Visualization Tool to Display Climatological Anomalies over Time and Space
No full textInternational audienceDetecting abnormal climate events across temporal and spatial scales is crucial to the understanding of local and regional climate trends. This demonstration introduces ClimBurst, a dynamic tool to detect climate bursts, which are unusually high or low values of one or more climate variables over some time interval. ClimBurst detects bursts without prior assumptions about their temporal duration. The demonstration will allow users to interact directly with our system to see both a summary showing the presence/absence of bursts over a user-specified year and spatial range. The demonstration will also allow users to perform time-travel queries to see how bursts propagate over space and time
QtNURAV: A Robust Latch Design with Quintuple Node Upset Recovery and Algorithm based Verifications for Aerospace Applications
No full textInternational audienceWith the continuous scaling of transistor feature sizes to the deep nano-scale, modern circuits have become increasingly sensitive to radiation-induced soft errors, such as multiple node upset (MNU). Especially, due to charge-sharing, the striking of one high-energy radiative particle can simultaneously impact multiple adjacent nodes of a highly integrated circuit causing MNU, e.g., quintuple node upset (QtNU) and even more node upsets. Existing MNU-hardened latch designs cannot provide QtNU recovery, and their robustness verification heavily depends on electronic-designautomation tools. In this paper, we present a robust latch, namely QtNURAV, to protect against QtNU for aerospace applications, along with an algorithm-based QtNU recovery verification method. The latch mainly employs four parallel dual-interlocked cells (DICEs) that feed each other through eight 3-input inverters to robustly retain values. The proposed algorithm simplifies the fault-tolerance verification process, demonstrating the QtNU recovery of QtNURAV. Simulation results based on HSPICE show that the QtNURAV latch can self-recover from any possible QtNU and has low delay and power consumption. Compared to existing Quadruple Node Upset hardened latches, the proposed QtNURAV latch provides QtNU recovery and effectively reduces delay, power and delay-power-area comprehensive product by 33.3%, 30.1%, and 53.3% on average, respectively
Ergodic theorem and algorithmic randomness (following V. V'yugin)
No full textInternational audienceIn this expository note we present the proof of the constructive version of Birkhoff's ergodic theorem following Vladimir V'yugin, trying to separate and state explicitly the combinatorial statement on which this proof is based. The exposition is based on his papers and explanations given during his visit to LIF (CNRS, Marseille)