HAL - Lille 3
Not a member yet
5967 research outputs found
Sort by
Automatic Fall Detection System using Sensing Floors
No full textInternational audienceAutomatic fall detection is a major issue in taking care of the health of elderly people and has the potentialof increasing autonomy and independence while minimizing the risks of living alone. It has been an active researcharea due to the large demand of the healthcare association for fall detection goods. Fortunately, due to the recentfast progression in sensing technologies, fall detection system becomes prospective. It permits to monitor elders anddetect their falls, and consequently provides emergency support whenever needed. This paper describes the currentwork of detecting falls in independent living apartments using accelerometer concealed under tiles. We present theup-to-date advancement of data collection, feature extraction, feature selection, and signal changing detection, whichare essential phases of this work
Traitement de signaux irrégulièrement échantillonnés issus du mobile crowd-sensing pour la surveillance de la qualité de l’air
No full textL'Internet des Objets (IoT pour Internet of Things en anglais) étend internet aux choses et aux lieux réels : les objets connectés peuvent alors communiquer entre eux mais aussi avec leurs utilisateurs. Un intérêt de l'IoT est de permettre (i) la production d'une grande quantité d'information (Big Data) par un réseau distribué d'objets connectés et en conséquence (ii) une meilleure analyse de phénomènes physiques ou de comportements. Parmi les méthodes d'acquisition de l'IoT, le mobile crowd-sensing fournit des données datées et géo-localisées, produites par une foule de capteurs mobiles (issus de ou connectés à des smartphones) et transmises à un serveur via un réseau sans fil, de type WiFi ou GSM. L'exploitation des données provenant des capteurs pour l'IoT nécessite le développement de méthodes spécifiques de traitement de l'information pour améliorer la confiance en leur qualité—c'est-à-dire, avoir des réponses cohérentes entre les capteurs, détecter les valeurs aberrantes ou les capteurs défaillants—qui ne peut pas être réalisé manuellement en laboratoire. Pour acquérir ces données, nous nous appuyons sur la plate-forme APISENSE® (http://apisense.io) qui facilite le déploiement et l’orchestration à grande échelle de collectes de données sur le terrain. En particulier, nous utilisons APISENSE® pour collecter et agréger des données de qualité de l’air récupérées par des modules Arduino avant d’être enrichies et transmises par des smartphones. Les données sont ensuite automatiquement fusionnées en ligne pour déterminer les paramètres d’étalonnage des différents capteurs et ainsi améliorer la qualité de leurs relevés. Le problème d’étalonnage à distance est revisité comme un problème informé de factorisation matricielle à données manquantes, où les facteurs contiennent respectivement le modèle d'étalonnage fonction du phénomène physique observé (qui peut être affine, multi-linéaire, ou non-linéaire) et les paramètres d'étalonnage de chaque capteur. Les approches proposées sont montrées plus performantes que des approches basées sur la complétion de la matrice de données observées. D'un point de vue applicatif, nous voulons coupler des mesures citoyennes—anonymisées—de la qualité de l’air aux mesures normalisées—très précises mais très parcimonieuses—de l'association régionale agréée pour la surveillance de la qualité de l'air, afin de reconstruire des cartes fines de la qualité de l'air à l'échelle d'un quartier. Les travaux en cours concernent notamment le déploiement des boîtiers de mesure open-source, fabriqués puis portés par des lycéens et des étudiants volontaires
Word2Vec vs DBnary: Augmenting METEOR using Vector Representations or Lexical Resources?
No full textInternational audienceThis paper presents an approach combining lexico-semantic resources and distributed representations of words applied to the evaluation in machine translation (MT). This study is made through the enrichment of a well-known MT evaluation metric: METEOR. This metric enables an approximate match (synonymy or morphological similarity) between an automatic and a reference translation. Our experiments are made in the framework of the Metrics task of WMT 2014. We show that distributed representations are a good alternative to lexico-semantic resources for MT evaluation and they can even bring interesting additional information. The augmented versions of METEOR, using vector representations, are made available on our Github page
Fuzzy unknown input observer-based robust fault estimation design for discrete-time fuzzy systems
No full textInternational audienc
Fault Tolerant Multi-Sensor Fusion for Multi-Robot Collaborative Localization
No full textInternational audienceMulti-robot system is used in some unreachable or dangerous area in order to replace the human operators. In such environments the integrity of localization should be assured by adding a sensor fault diagnosis step. In this paper, we present a method able, in addition of localizing a group of robots, to detect and exclude the faulty sensors from the team. The estimator is the informational form of the Kalman Filter (KF) namely Information Filter (IF). The developed residual test is based on the divergence between the predicted and the corrected estimation of the IF, calculated in term of the Kullback-Leibler divergence (KLD). The main contributions of this paper: - developing a method able simultaneously to localize a group of robots and to detect the faulty sensors - using the IF and the KLD as a residual test - Application of the proposed framework to a real environment with real robots
Production-Driven Patch Generation and Validation
No full textWe envision a world where the developer would receive each morning in her GitHub dashboard a list of potential patches that fix certain production failures. For this, we propose a novel program repair scheme, with the unique feature of being applicable to production directly. We present the design and implementation of a prototype system for Java, called Itzal, that performs patch generation for uncaught exceptions in production. We have performed two empirical experiments to validate our system: the first one on 34 failures from 14 different software applications, the second one on 16 seeded failures in 3 real open-source e-commerce applications for which we have set up a realistic user traffic. This validates the novel and disruptive idea of using program repair directly in production
Design Patterns pour les environnements dans les simulations multi-agents
No full textInternational audienceEnvironment, usually regarded as one of the key concepts of MAS especially in simulation, is however rarely specified in a precise or even explicit way, since its implementation is assumed obvious or given. On the contrary, we argue that the way of modeling space and connections between agents in a simulation, allows only a few efficient implementation solutions. We aim at formalizing the fundamental purposes of the environment, i.e. helping the agents to find their neighbors, and providing them with information. Thus, the search for a balance between modeling issues on the one hand (environment topology, nature of the information) and the operational priorities on the other hand (execution efficiency, relevance of knowledge representation), outlines four environment patterns. Through this unifying approach, the usual, monolithical and sometimes complex, "environment" of a multiagent simulation can be modeled and implemented as the combination of severals patterns.L'environnement, considéré généralement comme un des concepts clefs des SMA, tout particulièrement en simulation, fait pourtant rarement l'objet d'une spécification précise ou même explicite, car son implémentation est considérée comme évidente ou donnée. Nous défendons au contraire l'idée que la façon dont on modélise l'espace ou les relations entre agents dans une simulation, conduit à la mise en œuvre d'un nombre réduit de solutions efficaces. Notre démarche vise à formaliser les fonctions fondamentales de l'environnement : permettre aux agents de localiser leurs voisins, et leur fournir de l'information. Ainsi, les compromis entre choix de modélisation (topologie de l'environnement, nature des informations) d'une part, et priorités opérationnelles (efficacité d'exécution, pertinence de la représentation des connaissances) d'autre part, permettent d'identifier quatre grand patterns d'environnements. Dans cette approche unificatrice, "l'environnement" habituellement monolithique et parfois complexe d'une simulation multi-agent peut être modélisé et implémenté comme la combinaison de plusieurs patterns
A Comparison Between Deep Neural Nets and Kernel Acoustic Models for Speech Recognition
No full textInternational audienceWe study large-scale kernel methods for acoustic modeling and compare to DNNs on performance metrics related to both acoustic modeling and recognition. Measuring perplexity and frame-level classification accuracy, kernel-based acoustic models are as effective as their DNN counterparts. However, on token-error-rates DNN models can be significantly better. We have discovered that this might be attributed to DNN's unique strength in reducing both the perplexity and the entropy of the predicted posterior probabilities. Motivated by our findings, we propose a new technique, entropy regularized perplexity, for model selection. This technique can noticeably improve the recognition performance of both types of models, and reduces the gap between them. While effective on Broadcast News, this technique could be also applicable to other tasks
A Supervised Approach for Rhythm Transcription Based on Tree Series Enumeration
No full textInternational audienceWe present a rhythm transcription system integrated in the computer-assisted composition environment OpenMusic. Rhythm transcription consists in translating a series of dated events into traditional music notation's pulsed and structured representation. As transcription is equivocal, our system favors interactions with the user to reach a satisfactory compromise between various criteria, in particular the precision of the transcription and the readability of the output score. It is based on a uniform approach, using a hierarchical representation of duration notation in the form of rhythm trees, and an efficient dynamic-programming algorithm that lazily evaluates the transcription solutions. It is run through a dedicated user interface allowing to interactively explore the solution set, visualize the solutions and locally edit them
Relationships between Consumption, Publication and Impact in French Universities in a value perspective : A Bibliometric Analysis
No full textInternational audienceThe study aims to investigate the relationships between consumption of e-journals distributed by Elsevier ScienceDirect platform, publication (articles) and impact (citations) in a sample of 13 French universities, from 2003 to 2009. It adopts a value perspective as it questions whether or not publication activity and impact are some kind of return led by consumption. A bibliometric approach was used to explore the relations between these three variables. The analysis developed indicators inspired by the mathematical h-Index technique. Results show that the relation between consumption, publication and citations depends on the discipline’s profile, the intensity of research and the size of each institution. Moreover, although relations have been observed between the three variables, it is not possible to determine which variable comes first to explain the phenomena. The study concludes by showing strong correlations, which nevertheless do not lead to clear causal relations. The article provide practical implication for academic library managers who want to show the added value of their electronic e-journals collections can replicate the study approach. Also for policy makers who want to take into account e-journals usage as an informative tool to predict the importance of publication activity. Originality: The study is the first French contribution to e-journal value studies. Its originality consists in developing a value viewpoint that relies on a bibliometric approach