Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
Combining thresholded real values for designing an artificial neuron in a neural network
No full textInternational audienceThis study emanates from a simple observation: as specified by Vapnik [37] in his study, an artificial neural network cannot generate a universal approximator if the aggregation function chosen to design the artificial neuron does not include non-linearity. The usual option is to follow a linear aggregation by a non-linear function, or so-called activation function. We wonder if this approach could be replaced by one using a natively non-linear aggregation function.Among all of the available non-linear aggregation functions, here we are interested in aggregations based on weighted minimum and weighted maximum operations [8]. As these operators were originally developed within a possibility theory and fuzzy rule framework, such operators cannot be easily integrated into a neural network because the values that are usually considered belong to [0, 1]. For gradient descent based learning, a neuron must be an aggregation function derivable with respect to its inputs and synaptic weights, whose variables (synaptic weights, inputs and outputs) must all be signed real values. We thus propose an extension of weighted maximum based aggregation to enable this learning process. We show that such an aggregation can be seen as a combination of four Sugeno integrals. Finally, we compare this type of approach with the classical one.</p
Disparity estimation of stereo-endoscopic images using deep generative network
No full textInternational audienceA novel disparity estimation pipeline is proposed for 3D reconstruction of dynamic soft tissues in minimally invasive surgery (MIS), which uses a deep generative network to learn manifold distributions of reasonable disparity maps from past stereo images in the training phase, and transforms stereo matching into an optimization problem with respect to the low-dimensional latent vector of the learned generator in the application phase. The proposed pipeline is particularly suitable for dynamic MIS scenarios with insufficient training data, as the photometric loss is explicitly used in the application phase and the scenario priors are introduced via a deep generative network
Event Detection in Time Series
No full textInternational audienceThis book is dedicated to exploring and explaining time series event detection in databases. The focus is on events, which are pervasive in time series applications where significant changes in behavior are observed at specific points or time intervals. Event detection is a basic function in surveillance and monitoring systems and has been extensively explored over the years, but this book provides a unified overview of the major types of time series events with which researchers should be familiar: anomalies, change points, and motifs. The book starts with basic concepts of time series and presents a general taxonomy for event detection. This taxonomy includes (i) granularity of events (punctual, contextual, and collective), (ii) general strategies (regression, classification, clustering, model-based), (iii) methods (theory-driven, data-driven), (iv) machine learning processing (supervised, semi-supervised, unsupervised), and (v) data management (ETL process). This taxonomy is weaved throughout chapters dedicated to the specific event types: anomaly detection, change-point, and motif discovery. The book discusses state-of-the-art metric evaluations for event detection methods and also provides a dedicated chapter on online event detection, including the challenges and general approaches (static versus dynamic), including incremental and adaptive learning. This book will be of interested to graduate or undergraduate students of different fields with a basic introduction to data science or data analytics
Digital-Based Solution for the Generation of FM/PM Test Stimuli
No full textInternational audienceThis paper explores a low-cost solution for generating modulated test stimuli using a standard digital Automated Test Equipment (ATE). The technique relies on the generation of a modulated binary signal with appropriate encoding using a digital tester channel and the exploitation of one of its harmonic replicas. A theoretical analysis is presented in this paper, considering a simple modulation scheme, i.e. single-tone frequency or phase modulation. The relationship between the baseband spectrum and the harmonic replicas is established and an analytical expression of the modulated digital signal is derived, taking into account effects associated with discrete-time generation. A corruption estimator is then defined, enabling nondestructive sampling conditions to be identified. Experimental results are provided demonstrating the ability of the proposed solution to generate a modulated signal with the desired characteristics at a frequency higher than that of the test equipment
Analysis and Perspectives on the ANA Avatar XPRIZE Competition
No full textInternational audienceThe ANA Avatar XPRIZE was a four-year competition to develop a robotic “avatar” system to allow a human operator to sense, communicate, and act in a remote environment as though physically present. The competition featured a unique requirement that judges would operate the avatars after less than one hour of training on the human–machine interfaces, and avatar systems were judged on both objective and subjective scoring metrics. This paper presents a unified summary and analysis of the competition from technical, judging, and organizational perspectives. We study the use of telerobotics technologies and innovations pursued by the competing teams in their avatar systems, and correlate the use of these technologies with judges’ task performance and subjective survey ratings. It also summarizes perspectives from team leads, judges, and organizers about the competition’s execution and impact to inform the future development of telerobotics and telepresence
Modélisation et évaluation d’un concept de mini data center à faible impact environnemental
No full textData centers are essential to today’s digital infrastructure, yet their environmental footprint remains a significant concern. They consume large amounts of energy, contribute to greenhouse gas emissions, and depend on finite natural resources. Additionally, their decommissioning generates electronic waste, further amplifying their ecological impact. To address these challenges, recent research has focused on improving energy efficiency and promoting eco-design. Energy efficiency strategies target hardware and software optimizations to reduce energy consumption, while eco-design integrates environmental considerations from the outset, aiming to minimize impact across the entire lifecycle, from equipment manufacturing to end-of-life. This includes reducing energy use, optimizing material consumption, and enhancing component recyclability or reuse. This thesis investigates an existing eco-design approach tailored for data centers : Genesis. The concept is built on a distributed network of "green" computing nodes that share both computational loads and energy. Each node includes a solar panel, a battery, a server, and an energy exchange system, allowing for greater autonomy and resilience. As with any renewable-powered infrastructure, accurate sizing of components (servers, solar panels, and batteries) is crucial to balance performance and cost, avoiding both overprovisioning and underprovisioning. Furthermore, Genesis must ensure continuous operation despite potential component failures. To meet these challenges, this work introduces a formal modeling framework based on timed automata, enabling rigorous analysis and optimization of resource allocation. The model supports efficient workload scheduling and strategic server renewal, helping to reduce both energy drawn from the grid and the consumption of materials required for equipment production, favoring greater use of local renewable energy. To evaluate the environmental benefits of Genesis, a comparative Life Cycle Assessment was conducted against a conventional data center design. This multi-criteria analysis focused on key environmental indicators : climate change, abiotic resource depletion, acidification, ionizing radiation, and particulate matter emissions. It also considered variables such as scalability, energy mix, and the extended lifespan of computing nodes. The findings show that Genesis substantially reduces environmental impacts across all five indicators, highlighting its potential as a foundation for more sustainable data center architectures.Les data centers occupent une place centrale dans l’infrastructure numérique moderne, mais leur empreinte environnementale demeure préoccupante. Leur fonctionnement requiert d’importantes quantités d’énergie, contribue aux émissions de gaz à effet de serre et mobilise des ressources naturelles limitées. Leur fin de vie engendre également des déchets électroniques, accentuant leur impact écologique. Pour y remédier, de nombreux travaux ont été menés ces dernières années, notamment autour de l’amélioration de l’efficacité énergétique et de l’éco-conception. Les premières approches visent à réduire la consommation des équipements par des optimisations matérielles et logicielles, tandis que l’éco-conception intègre les enjeux environnementaux dès la phase de conception, en cherchant à limiter l’impact tout au long du cycle de vie du data center, de la fabrication au démantèlement. Dans ce contexte, cette thèse s’appuie sur une approche d’éco-conception existante : Gene- sis. Cette méthode repose sur un réseau distribué de nœuds informatiques « verts » alimentés par énergie solaire. Chaque nœud intègre un panneau solaire, une batterie, un serveur et un système de partage d’énergie, permettant une mutualisation des ressources computationnelles et énergétiques. L’objectif est de concevoir des data centers sobres et résilients, capables de maintenir leur fonctionnement malgré des pannes ou des fluctuations de production énergétique. L’un des enjeux majeurs réside dans le dimensionnement optimal des ressources (serveurs, panneaux solaires, batteries), afin d’éviter à la fois le surdimensionnement coûteux et le sous-dimensionnement pénalisant. La thèse propose un cadre de modélisation basé sur les automates temporisés, permettant d’analyser et d’optimiser cette allocation de ressources de façon rigoureuse. Ce modèle intègre également des stratégies de planification des charges de travail et de renouvellement des serveurs, afin de réduire à la fois la consommation énergétique et l’usage de matériaux nécessaires à la fabrication des équipements. Pour évaluer l’impact environnemental global de l’approche Genesis, une analyse de cycle de vie multicritère a été conduite, comparant Genesis à une architecture de data center classique. L’analyse porte sur plusieurs indicateurs environnementaux : changement climatique, épuisement des ressources abiotiques, acidification, radiations ionisantes et émissions de particules. Elle examine aussi l’influence du passage à l’échelle, du mix énergétique, et de la durée de vie des équipements. Les résultats montrent une réduction significative de l’empreinte environnementale avec Genesis selon ces critères, confirmant le potentiel de cette approche pour concevoir des data centers durables
A comparison of hybrid maps for the visualization of two geolocated quantitative variables
No full textInternational audienceThematic maps, like choropleth maps, symbol maps or cartograms, are commonly used to visualize spatial quantitative data. Many studies have been conducted to compare the different approaches and thus define the best strategies to produce suitable and efficient maps. When analyzing spatial data, it is also often necessary to visualize and compare several variables on the same map. Therefore, the question arises of how to best associate two variables in a single representation without one of them prevailing over the other, while avoiding overloading the map and making it difficult to interpret. In this article, we propose a comparison of five types of bivariate maps based on a user study. Participants performed a set of tasks using different maps produced from multiple datasets. Our analysis is based on three approaches: (1) quantitative analysis of user answer accuracy, (2) quantitative analysis of user answer times, and (3) quantitative and qualitative analysis of user feedback. The results suggest that combining symbol and choropleth maps is the most effective approach among those tested, while combining cartograms with any technique is the worst
Humanoid-Human Sit-to-Stand-to-Sit Assistance
No full textInternational audienceStanding and sitting are basic tasks that become increasingly difficult with age or frailty. Assisting these movements using humanoid robots is a complex challenge, particularly in determining where and how much force the robot should apply to effectively support the human's dynamic motions. In this letter, we propose a method to compute assistive forces directly from the human's dynamic balance, using criteria typically employed in humanoid robots. Specifically, we map humanoid dynamic balance metrics onto human motion to calculate the forces required to stabilize the human's current posture. These forces are then applied at the appropriate locations on the human body by the humanoid. Our approach combines the variable height 3D divergent component of motion with gravito-inertial wrench cones to define a 3D balance region. Using centroidal feedback, we compute the required assistance force to maintain balance and distribute the resulting wrenches across the human's body using a humanoid robot dynamically balanced according to the same criteria. We demonstrate the effectiveness of this framework through both simulations and experiments, where a humanoid assists a person in sit-to-stand and stand-to-sit motions, with the person wearing an age-simulation suit to emulate frailty
Can Masked Autoencoders Also Listen to Birds?
No full textInternational audienceMasked Autoencoders (MAEs) learn rich representations in audio classification throughan efficient self-supervised reconstruction task. Yet, general-purpose models struggle infine-grained audio domains such as bird sound classification, which demands distinguishingsubtle inter-species differences under high intra-species variability. We show that bridgingthis domain gap requires full-pipeline adaptation beyond domain-specific pretraining data.Using BirdSet, a large-scale bioacoustic benchmark, we systematically adapt pretraining,fine-tuning, and frozen feature utilization. Our Bird-MAE sets new state-of-the-art resultson BirdSet’s multi-label classification benchmark. Additionally, we introduce the parameter efficient prototypical probing, which boosts the utility of frozen MAE features by achieving up to 37 mAP points over linear probes and narrowing the gap to fine-tuning in low-resource settings. Bird-MAE also exhibits strong few-shot generalization with prototypical probes on our newly established few-shot benchmark on BirdSet, underscoring the importance of tailored self-supervised learning pipelines for fine-grained audio domains
The point is the mask: scaling coral reef segmentation with weak supervision
No full textInternational audienceMonitoring coral reefs at large spatial scales remains an open challenge, essential for assessing ecosystem health and informing conservation efforts. While drone-based aerial imagery offers broad spatial coverage, its limited resolution makes it difficult to reliably distinguish fine-scale classes, such as coral morphotypes. At the same time, obtaining pixel-level annotations over large spatial extents is costly and labor-intensive, limiting the scalability of deep learning-based segmentation methods for aerial imagery. We present a multi-scale weakly supervised semantic segmentation framework that addresses this challenge by transferring fine-scale ecological information from underwater imagery to aerial data. Our method enables large-scale coral reef mapping from drone imagery with minimal manual annotation, combining classification-based supervision, spatial interpolation and self-distillation techniques. We demonstrate the efficacy of the approach, enabling large-area segmentation of coral morphotypes and demonstrating flexibility for integrating new classes. This study presents a scalable, cost-effective methodology for high-resolution reef monitoring, combining low-cost data collection, weakly supervised deep learning and multi-scale remote sensing