Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
Accuracy-Power Trade-Offs in GNSS-Based Positioning for Frugal Floating Devices
No full textInternational audienceThis paper quantifies the accuracy versus power trade-off in ultra low-power GNSS tags for marine tracking. Field tests (more than 22000 fixes) on land and in a coastal lagoon reveal that receiver metadata (fix quality and 2D/3D status) outperform EHPE as predictors of true error. Building on these insights, we propose an embedded fix selection algorithm that stops acquisition once spatial stability and quality thresholds are met. For field deployment conditions, the method cuts energy use by about 40 percent while keeping median error within 5 m of the best available fix, extending tag autonomy from 9 to 15 days without hardware changes
The -Binding Function of -Directional Segment Graphs
No full textInternational audienceAbstract Given a positive integer d , the class d -DIR is defined as all those intersection graphs formed from a finite collection of line segments in having at most d slopes. Since each slope induces an interval graph, it easily follows for every G in d -DIR with clique number at most that the chromatic number of G is at most . We show for every even value of how to construct a graph in d -DIR that meets this bound exactly. This partially confirms a conjecture of Bhattacharya, Dvořák and Noorizadeh. Furthermore, we show that the -binding function of d -DIR is for even and for odd. This extends an earlier result by Kostochka and Nešetřil, which treated the special case
How a task-blind adaptive VR system can improve users' task performance: an assisted immersive analytics use case
No full textInternational audienceRecently, some works have built adaptive systems providing assistance to the user in virtual reality (VR), with little or no knowledge of the user’s task. These task-blind help systems can influence behaviours and exploration strategies; however, their ability to significantly improve users’ performance on their tasks is still unclear. In this study, we aim to clarify the impact of task-blind help systems on user performance. We also explore two avenues that could provide a better understanding of why these systems can be effective and interesting to study. Our controlled user study involved 56 participants in an immersive analytics environment and compared four VR help-system configurations, including three task-blind systems and a no-assistance baseline. Results showed significant task performance improvements with one task-blind system, highlighting user control as a key factor of efficiency. This work demonstrates the potential of task-blind help systems, offering a flexible framework for adaptive design and raising questions about their broader applications
OntoPFAS : Ontologie des PFAS et de leur exposition
No full textMISTEA - Axe informatiqueInternational audienceLa construction d'ontologies est une des premières tâches dans le domaine de la représentation des connaissances. Elle reste très pertinente aujourd'hui, grâce à l'expressivité des langages formels, qui permettent encore d'explorer et de découvrir des connaissances. Pendant des décennies, la communauté a développé des méthodologies pour la construction manuelle d'ontologies, et plusieurs classifications de celles-ci ont été proposées. Dans cet article, nous présentons une méthodologie de construction d'ontologie basée sur des méthodes existantes, et nous l'appliquons à la représentation du domaine des PFAS (Per-et poly-fluoroalkyle substances) et de leur exposition. Les PFAS sont des substances dont la structure chimique particulière les rend très résistantes et efficaces dans de nombreuses applications industrielles. Ils suscitent un intérêt croissant en raison de leur impact négatif sur la santé et l'environnement. Ce travail s'inscrit dans le cadre du projet interdisciplinaire DAE (Détection d'Anomalies Environnementales)
Class conditional conformal prediction for multiple inputs by p-value aggregation
No full textInternational audienceConformal prediction methods are statistical tools designed to quantify uncertainty and generate predictive sets with guaranteed coverage probabilities. This work introduces an innovative refinement to these methods for classification tasks, specifically tailored for scenarios where multiple observations (multi-inputs) of a single instance are available at prediction time. Our approach is particularly motivated by applications in citizen science, where multiple images of the same plant or animal are captured by individuals. Our method integrates the information from each observation into conformal prediction, enabling a reduction in the size of the predicted label set while preserving the required class-conditional coverage guarantee. The approach is based on the aggregation of conformal p-values computed from each observation of a multi-input. By exploiting the exact distribution of these p-values, we propose a general aggregation framework using an abstract scoring function, encompassing many classical statistical tools. Knowledge of this distribution also enables refined versions of standard strategies, such as majority voting. We evaluate our method on simulated and real data, with a particular focus on Pl@ntNet, a prominent citizen science platform that facilitates the collection and identification of plant species through user-submitted images
Generating realistic artificial human genomes using adversarial autoencoders
No full textInternational audienceA publicly available human genome is both valuable to researchers and a risk for its donor. Many actors could exploit it to extract information about the donor’s health or that of their relatives. Recent efforts have employed artificial intelligence models to simulate genomic data, aiming to create synthetic datasets with scientific merit while preserving patient anonymity. Challenges arise due to the vast amount of data that constitute a complete human genome and the computational resources required. We present a dimension reduction method that combines artificial intelligence with our knowledge of in vivo mutation association mechanisms. This approach enables processing large amounts of data without significant computational resources. Our genome segmentation follows chromosomal recombination hotspots, closely resembling mutation transmission mechanisms. Data from the 1000 Genomes Project are used to train variational autoencoders with a Wasserstein GAN to generate novel data in a two-step process. After optimizing our strategy, our pipeline can generate a simulated population meeting several essential criteria. They are diverse but realistic; the newly generated combinations of mutations follow linkage disequilibrium found in humans. Our pipeline does not reveal the genetic identity of any individual donor, synthesizing genomes that differ from reference samples
Comparaison de quatre modélisations des préférences
No full textInternational audienceDe nombreux systèmes d'aide à la décision multicritères sont basés sur l’apprentissage, à partir d’une base d’exemples issus d’experts, de fonctions dédiées associant à l’évaluation d’un ensemble de critères une note mimant la décision de ces experts.Les paramètres de cette fonction sont alors appris par régression grâce à une base d’exemples. Toutefois, il est souvent plus fiable de demander à un expert d'exprimer une préférence entre deux exemples plutôt qu'une note globale pour chaque exemple, surtout lorsque les exemples arrivent dans un flux séquentiel. Dans cet article, nous proposons de modéliser la relation de préférence par une fonction d’ensemble appelée « jeu » qui permet de répercuter, sur la sortie du système, non seulement l’influence de chaque critère mais aussi celle de coalitions de critères. Nous proposons d’apprendre les paramètres de cette fonction d’ensemble par régression.Quatre jeux différents sont étudiés : un jeu général, un jeu 2-additif, un jeu linéaire et un jeu maxitif. Des expériences sont menées avec des ensembles de données généralement utilisées pour tester des méthodes de régression
Justified Preference Aggregation in Agri-Food Systems: An approach, Argumentation Methods and Tools (Supplementary Material)
No full textInternational audienc
UniTED: A Unified Time Series Event Detection Repository
No full textInternational audienceEvent detection in time series is essential for numerous real-world applications, from monitoring industrial systems to identifying health anomalies. Public annotated datasets are crucial for benchmarking, training, and validating detection models. Despite recent advances in the field, there is a lack of a standardized and unified repository for evaluating different event types, which limits progress in reproducibility, comparability, and model development. This paper presents the UniTED, a Unified Event Detection Dataset for time series. UniTED consolidates annotated series from diverse domains and offers a common format and protocol for evaluation. The repository supports three event types: anomalies, change points, and motifs. UniTED fosters reusability and reproducibility, contributing to improved performance assessment and model generalization across data analysis tasks. However, existing datasets have limitations, including poor standardization, a lack of annotation guidelines, limited support for different event types, and difficulties in automating performance evaluation. UniTED presents a harmonized ETL process, label and annotation conventions, and an open-source implementation. Three use cases are presented to demonstrate the applicability of the dataset
Two Arms, One Goal: Reproducible Dual-Arm Robotic Fruit Harvesting
No full textInterest in agricultural robotics has significantly increased, due to the potential benefits of improved productivity and labor reduction. Yet, developing robotic harvesters for unstructured environments presents many challenges for robot perception, planning and action. Here, we propose a dual-arm approach to fruit harvesting. Our dual-arm fruit harvesting robots are equipped with an RGB-D camera, a cutting and a collecting tool. The vision system combines fruit detection and tracking. We detect branches and the tree trunk as obstacles, to ensure efficient and safe harvesting. To speed up the process, we implement a scheduling algorithm based on the traveling salesperson problem (TSP) to minimize the distance required to harvest all visible fruits. Finally, we use global motion planning to ensure collision-free operation. We validate our methods using two distinct dual-arm robots: Baxter, a cost-effective and widely available option, and BAZAR, a custom high precision dualarm platform. Furthermore, our experiments are conducted on two types of fruit: apples and oranges. In lab trials, the systems harvested 5 oranges and 6 apples, with average perfruit times of 20.6 s and 8.2 s and no collisions. Perception achieved mAP@50 = 0.886 on a 3,600-image dataset, MOTA = 89.8% with 3 ID switches, and 3D localization MAE of 7.8-21.1 mm. The results demonstrate the effectiveness of our dual-arm approach increasing adaptability of robotic fruit harvesting. The results demonstrate the effectiveness of our dual-arm approach in enhancing the efficiency and adaptability of robotic fruit harvesting