Search

Registered as Defensive Paper of the Intellectual Property Invention "Clustering Techniques for Revenue Accounting Error-Handling Automation", Patent number ID2326WW00, Intellectual Property Board, Amadeus S.A.S., Sophia Antipolis, France. July, 2020.; International audience; Semi-supervised consensus clustering integrates supervised information into consensus clustering in order to improve the quality of clustering. In this paper, we study the novel Semi-MultiCons semi-supervised consensus clustering method extending the previous MultiCons approach. Semi-MultiCons aims to improve the clustering result by integrating pairwise constraints in the consensus creation process and infer the number of clusters K using frequent closed itemsets extracted from the ensemble members. Experimental results show that the proposed method outperforms other state-of-art semi-supervised consensus algorithms.

Part 3: Recommendation System; International audience; Vector representations learning (also known as embeddings) for users (items) are at the core of modern recommendation systems. Existing works usually map users and items to low-dimensional space to predict user preferences for items and describe pre-existing features (such as ID) of users (or items) to obtain the embedding of the user (or item). However, we argue that such methods neglect the dual role of users, side information of users and items (e.g., dual citation relationship of authors, authoritativeness of authors and papers) when recommendation is performed for scientific paper. As such, the resulting representations may be insufficient to predict optimal author citations.In this paper, we contribute a new model named scientific paper recommendation using Author’s Dual Role Citation Relationship (ADRCR) to capture authors’ citation relationship. Our model incorporates the reference relation between author and author, the citation relationship between author and paper, and the authoritativeness of authors and papers into a unified framework. In particular, our model predicts author citation relationship in each specific class. Experiments on the DBLP dataset demonstrate that ADRCR outperforms state-of-the-art recommendation methods. Further analysis shows that modeling the author’s dual role is particularly helpful for providing recommendation for sparse users that have very few interactions.

International audience; The 4th International Workshop on Knowledge Discovery in Inductive Databases (KDID 2005) was held in Porto, Portugal, on October 3, 2005 in conjunction with the 16th European Conference on Machine Learning and the 9th European Conference on Principles and Practice of Knowledge Discovery in Databases. Ever since the start of the ?eld of data mining, it has been realized that the integration of the database technology into knowledge discovery processes was a crucial issue. This vision has been formalized into the inductive database perspective introduced by T. Imielinski and H. Mannila (CACM 1996, 39(11)). The main idea is to consider knowledge discovery as an extended querying p- cess for which relevant query languages are to be speci?ed. Therefore, inductive databases might contain not only the usual data but also inductive gener- izations (e. g. , patterns, models) holding within the data. Despite many recent developments, there is still a pressing need to understand the central issues in inductive databases. Constraint-based mining has been identi?ed as a core technology for inductive querying, and promising results have been obtained for rather simple types of patterns (e. g. , itemsets, sequential patterns). However, constraint-based mining of models remains a quite open issue. Also, coupling schemes between the available database technology and inductive querying p- posals are not yet well understood. Finally, the de?nition of a general purpose inductive query language is still an on-going quest.

ObjectiveTo investigate spatial heterogeneity of stunting prevalence among children in Côte d’Ivoire and examine changes in stunting between 1994 and 2011, to assess the impact of the 2002–2011 civil war that led to temporary partitioning of the country.DesignData from 1994, 1998 and 2011 Côte d’Ivoire Demographic and Health Surveys (DHS) were analysed using a geostatistical approach taking into account spatial autocorrelation. Stunting data were interpolated using ordinary kriging; spatial clusters with high and low stunting prevalence were identified using Kulldorff spatial scan statistics. Multilevel multivariable logistic regression was then carried out, with year of survey as the main independent variable and an interaction term for time by geographic zone (Abidjan, South, North).SettingCôte d’Ivoire, West Africa.SubjectsChildren aged 0–35 months included in three DHS (n6709).ResultsOverall stunting prevalence was 30·7, 28·7 and 27·8 % in 1994, 1998 and 2011, respectively (P=0·32). Clusters with high prevalence were found in 1994 (in the West region,PPP=0·01, respectively), but not in 2011. Abidjan was included in a cluster with low prevalence in all surveys (PConclusionsIn Côte d’Ivoire, significant changes in stunting prevalence were observed at the sub-national level between 1994 and 2011.

International audience; Cell biologists have long aimed at quantitatively modeling cell function. Recently, the outstanding progress of high-throughput measurement methods and data processing tools has made this a realistic goal. The aim of this paper is twofold: First, to suggest that, while much progress has been done in modeling cell states and transitions, current accounts of environmental cues driving these transitions remain insufficient. There is a need to provide an integrated view of the biochemical, topographical and mechanical information processed by cells to take decisions. It might be rewarding in the near future to try to connect cell environmental cues to physiologically relevant outcomes rather than modeling relationships between these cues and internal signaling networks. The second aim of this paper is to review exogenous signals that are sensed by living cells and significantly influence fate decisions. Indeed, in addition to the composition of the surrounding medium, cells are highly sensitive to the properties of neighboring surfaces, including the spatial organization of anchored molecules and substrate mechanical and topographical properties. These properties should thus be included in models of cell behavior. It is also suggested that attempts at cell modeling could strongly benefit from two research lines: (i) trying to decipher the way cells encode the information they retrieve from environment analysis, and (ii) developing more standardized means of assessing the quality of proposed models, as was done in other research domains such as protein structure prediction.

Trees are among the most studied data structures and several techniques have consequently been developed for comparing two trees belonging to the same category. Until the end of year 2020, there was a serious lack of suitable metrics for comparing two weighted trees or two trees from different categories. The problem of comparing two tree sets was not also specifically addressed. These limitations have been overcome in a paper published in 2021 where a customizable metric based on hidden Markov models have been proposed for comparing two tree sets, each containing a mixture of trees belonging to various categories. Unfortunately, that metric does not allow the use of non metric-dependent classifiers which take descriptor vectors as inputs. This paper addresses this drawback by deriving a descriptor vector for each tree set from the behavior of its associated models after a sufficiently long time. The comparison between two tree sets is then realized by comparing their associated descriptor vectors. Classification experiments carried out on the databases FirstLast-L (FL), FirstLast-LW (FLW) and Stanford Sentiment Treebank (SSTB) respectively showed best accuracies of 99.75%, 99.75% and 87.22%. These performances are respectively 40.75% and 20.52% better than the tree Edit distance respectively for FLW and SSTB. Additional clustering experiments exhibited 54.25%, 98.75% and 75.53% of correctly clustered instances respectively for FL, FLW and SSTB. No clustering was performed in existing work.; Les arbres font partie des structures de données les plus étudiées et plusieurs techniques ont par conséquent été développées pour comparer deux arbres appartenant à la même catégorie. Jusqu'à la fin de l'année 2020, il y avait un sérieux manque de métriques appropriées pour comparer deux arbres pondérés ou deux arbres de catégories différentes. Le problème de la comparaison de deux ensembles d'arbres n'a pas non plus été spécifiquement abordé. Ces limites ont été surmontées dans un article publié en 2021 dans lequel une métrique paramétrable basée sur des modèles de Markov cachés a été proposée pour comparer deux ensembles d'arbres, chacun contenant un mélange d'arbres appartenant à différentes catégories. Malheureusement, cette métrique ne permet pas l'utilisation de classificateurs ne dépendant pas d'une mesure de distance qui prennent généralement des vecteurs en entrées. Le présent article s'attaque à cette limite en proposant un descripteur vectoriel pour chaque ensemble d'arbres à partir du comportement sur une longue période des modèles qui lui sont associés. La comparaison entre deux ensembles d'arbres est alors réalisée en comparant leurs descripteurs vectoriels respectifs. Des expériences en classification menées sur les bases de données FirstLast-L (FL), FirstLast-LW (FLW) et Stanford Sentiment Treebank (SSTB) ont respectivement permis d'obtenir les meilleures performances de 99,75 %, 99,75 % et 87,22 %. Ces performances sont respectivement 40,75% et 20,52% meilleures que la 'Tree Edit distance' respectivement pour FLW et SSTB. Des expériences supplémentaires en clustering ont permis d'avoir respectivement 54,25 %, 98,75 % et 75,53 % instances correctement groupées pour FL, FLW et SSTB. Aucun clustering n'avait été effectué dans les travaux existants.

International audience; The paper presents the cartographic processing of the Landsat TM image by the two unsupervised classification methods of SAGA GIS: ISODATA and K-means clustering. The approaches were tested and compared for land cover type mapping. Vegetation areas were detected and separated from other land cover types in the study area of southwestern Iceland. The number of clusters was set to ten classes. The processing of the satellite image by SAGA GIS was achieved using Imagery Classification tools in the Geoprocessing menu of SAGA GIS. Unsupervised classification performed effectively in the unlabeled pixels for the land cover types using machine learning in GIS. Following an iterative approach of clustering, the pixels were grouped in each step of the algorithm and the clusters were reassigned as centroids. The paper contributes to the technical development of the application of machine learning in cartography by demonstrating the effectiveness of SAGA GIS in remote sensing data processing applied for vegetation and environmental mapping.

International audience; In this paper, we propose an efficient KNN service, called KPS (KNN-Peer-Sampling). The KPS service can be used in various contexts e.g. recommendation systems, information retrieval and data mining. KPS borrows concepts from P2P gossip-based clustering protocols to provide a localized and efficient KNN computation in large-scale systems. KPS is a sampling-based iterative approach, combining ran-domness, to provide serendipity and avoid local minimum, and clustering , to ensure fast convergence. We compare KPS against the state of the art KNN centralized computation algorithm NNDescent, on multiple datasets. The experiments confirm the efficiency of KPS over NNDescent: KPS improves significantly on the computational cost while converging quickly to a close to optimal KNN graph. For instance, the cost, expressed in number of pairwise similarity computations, is reduced by ≈ 23% and ≈ 49% to construct high quality KNN graphs for Jester and MovieLens datasets, respectively. In addition, the randomized nature of KPS ensures eventual convergence, not always achieved with NNDescent.

International audience; We propose a multidimensional statistical analysis procedure using projection and clustering methods in order to identify coherent clusters in a set of lexical areas. The methodology includes a geographical factor, such as administrative divisions or land cover features, to help the identification of clusters. By applying this method on data from the Occitan language area in the south of France, we are able to identify new spatial patterns and lexical boundaries that do not match traditional dialect boundaries. Our method helps to suggest possible explanations for these new patterns.

International audience; The influence of authors is mostly based on their capacity to form specific self-contained and/or active research communities or topics while also inspiring fruitful spin-off research derived from those communities or topics. Accurately estimating author influence and its indirect effect in inspiring external research creativity must thus be based on very precise community roles as well as topic-role analysis on citing material. In this paper we thus implement, compare, and combine two different kinds of knowledge mapping approaches used to track authors’ influence over time—a recent topic-based mapping approach and an original hybrid community-based mapping approach. The experimental context of our study is the analysis of the dynamics and of the influence of the research of Prof. Liu Zeyuan, the most important contributor in the field of Science of Science in China. This analysis focuses on papers citing Professor Liu Zeyuan's research and highlights the full extent of his creative thinking.

International audience; This paper provides a short overview of space time series clustering, which can be generally grouped into three main categories such as: hierarchical, partitioning-based, and overlapping clustering. The first hierarchical category is to identify hierarchies in space time series data. The second partitioning-based category focuses on determining disjoint partitions among the space time series data, whereas the third overlapping category explores fuzzy logic to determine the different correlations between the space time series clusters. We also further describe solutions for each category in this paper. Furthermore, we show the applications of these solutions in an urban traffic data captured on two urban smart cities (e.g., Odense in Denmark and Beijing in China).The perspectives on open questions and research challenges are also mentioned and discussed that allow to obtain a better understanding of the intuition, limitations, and benefits for the various space time series clustering methods. This work can thus provide the guidances to practitioners for selecting the most suitable methods for their used cases, domains, and applications.

International audience; Sargassum seaweed forms large floating mats drifting on the oceans. These mats are increasingly beaching on Caribbean islands, threatening the local wildlife and economies. This paper focuses on their tracking from space, in order to monitor these mats. More specifically, we focus on clustering sargassum mats on satellite images. This constitutes an important building block of the sargassum monitoring system, which then predicts the drift of those clusters to anticipate beachings and warn the local authorities. The difficulty of the clustering operation comes from the noisy nature of input data: image artefacts, partial cloud occlusion, mats discontinuities due to sea conditions, and the difficulty of acquiring ground truth data. This paper details our approach to overcome those challenges. We propose a method (hereafter named Sargassum Mats Detection Method-SMDM) that improves the mats identification by combining a first artefact detection step, a tailored clustering algorithm and a region growing algorithm.

International audience; Clustering is impacted by the regular increase of sample sizes which provides opportunity to reveal information previously out of scope. However, the volume of data leads to some issues related to the need of many computational resources and also to high energy consumption. Resorting to binned data depending on an adaptive grid is expected to give proper answer to such green computing issues while not harming the quality of the related estimation. After a brief review of existing methods, a first application in the context of univariate model-based clustering is provided, with a numerical illustration of its advantages. Finally, an initial formalization of the multivariate extension is done, highlighting both issues and possible strategies.

International audience; In this paper, we use tools of information geometry to compare, average and classify histograms. Beta distributions are fitted to the histograms and the corresponding Fisher information geometry is used for comparison. We show that this geometry is negatively curved, which guarantees uniqueness of the notion of mean, and makes it suitable to classify histograms through the popular K-means algorithm. We illustrate the use of these geometric tools in supervised and unsupervised classification procedures of two medical data-sets, cardiac shape deformations for the detection of pulmonary hypertension and brain cortical thickness for the diagnosis of Alzheimer's disease.

In this paper, we propose a multivariate Hawkes framework for modelling and predicting cyber attacks frequency. The inference is based on a public data set containing features of data breaches targeting the US industry. As a main output of this paper, we demonstrate the ability of Hawkes models to capture self-excitation and interactions of data breaches depending on their type and targets. In this setting, we detail prediction results providing the full joint distribution of future cyber attacks times of occurrence. In addition, we show that a non-instantaneous excitation in the multivariate Hawkes model, which is not the classical framework of the exponential kernel, better fits with our data. In an insurance framework, this study allows to determine quantiles for number of attacks, useful for an internal model, as well as the frequency component for a data breach guarantee.

International audience; In this paper, we present a collaborative multi-agent based system for data mining. We have used two data mining model functions, clustering of variables in order to build homogeneous groups of attributes, association rules inside each of these groups and a multi-agent approach to integrate the both data mining techniques. For the association rules extraction, we use both apriori algorithm and genetic algorithm.The main goal of this paper is the evaluation of the association rules obtained by running apriori and genetic algorithm using quantitative datasets in multi agent environment.

This paper is related to a wide project aiming at discoving from different streams of information (i.e. daily publication from the Internet), weak signals possibly sent by whistleblowers. The current study presented in this paper tackles the particular problem of clustering topics at multi-levels from multiple documents, and then extracting meaningful descriptors, such as weighted lists of words. In this context, we present a novel idea combining LDA (in charge clustering) and Word2vec (providing a consistency metric regarding the partitioned topics) as potential method for limiting the "a priori" number of cluster K usually needed in classical partitioning approaches. We proposed 2 implementations of this idea, respectively able to: (1) finding the optimal K for LDA; (2) gathering the optimal clusters from different levels of clustering., L'étude présentée s'inscrit dans le cadre du développement d'une plateforme d'analyse automatique de documents associée à un service sécurisé lanceurs d'alerte, de type GlobalLeaks, focalisé sur la révélation de faits/événements/actions en lien avec des problématiques environnementales. Le présent article aborde le problème particulier du regroupement de sujets à plusieurs niveaux à partir de documents multiples, puis de l'extraction de descripteurs significatifs, tels que des listes pondérées de mots. Dans ce contexte, nous présentons une nouvelle idée combinant LDA (en charge du clustering) et Word2vec (fournissant une métrique de cohérence concernant les sujets partitionnés) comme méthode potentielle pour limiter le nombre "a priori" de cluster K habituellement nécessaire dans les approches classiques du partitionnement. Nous avons proposé 2 mises en œuvre de cette idée, respectivement en mesure de : (1) trouver le K optimal pour LDA ; (2) rassembler les clusters optimaux de différents niveaux de clustering.

International audience; We develop a fast-deployed crop health monitoring system using state-of-the-art technologies to collect data from crop fields in order to take appropriate and timely actions. For the proposed resource optimized system, Saudi Arabian agriculture is taken as a case study. To achieve the desire goals, we harness IoT (Internet of Things) and drones in agriculture to establish rapid system deployment. This paper focuses on the data collection from crop field by organizing heterogeneous IoT devices in clusters and localise them for data harvesting. Clusters are formed by considering the path of UAVs, sensors heterogeneity, weather conditions, fluctuation of sensor nodes, and the communication cost of IoT devices. For localisation, carrying larger or heavier arrays of antennas and receivers with a small size UAV is also a major issue considered in this paper. Hence, we introduce a dynamic clustering and virtual antenna array to develop a complete data collection scheme supported by simulations and experimental tests with proof-of-concept devices. The results are analysed and found promising in terms of energy efficiency, throughput, ease of use, and deployment time. Whole the system is developed with the concept that it can install in rural and remote area with minimum deployment time and agile enough that can collect data in worst conditions (bad weather, hostile environment, fluctuating nodes, poor infrastructure, with or without an established network). In broader sense it can map easily in many similar applications where data is needed to be harvested from a wide range of heterogeneous sensors without existing any infrastructure and ground topology.

International audience; This paper is related to a project aiming at discovering weak signals from different streams of information, possibly sent by whistleblowers in a platform as GlobalLeaks. The study presented in this paper tackles the particular problem of clustering topics at multi-levels from multiple documents, and then extracting meaningful descriptors, such as weighted lists of words for document representations in a multi-dimensions space. In this context, we present a novel idea which combines Latent Dirichlet Allocation and Word2Vec (providing a consistency metric regarding the partitioned topics) as potential method for limiting the "a priori" number of cluster k usually needed in classical partitioning approaches. We proposed 2 implementations of this idea, respectively able to: (1) finding the best k for LDA in terms of topic consistency; (2) gathering the optimal clusters from different levels of clustering. We also proposed a non-traditional visualization approach based on a multi-agents system which combines both dimension reduction and interactivity.; L'étude présentée s'inscrit dans le cadre du développement d'une plateforme d'analyse automatique de documents associée à un service sécurisé lanceurs d'alerte, de type GlobalLeaks. Cet article se focalise principalement sur la recherche de signaux faibles présents dans les documents. Il s'agit d'une problématique investiguée dans un grand nombre de champs disciplinaires et de cadres applicatifs. Nous supposons que chaque document est un mélange d'un petit nombre de thèmes ou catégories, et que la création de chaque mot est attribuable en termes de probabilités à l'un des thèmes du document. Les catégories des documents transmis ne sont pas connues"a priori". Les mots-clés présents dans les documents représentatifs de ces catégories sont également inconnus. L'analyse des documents reçus doit simultanément permettre de découvrir les thèmes, classer les documents relativement à ces thèmes, détecter les mots-clés pertinents relatifs aux thèmes et enfin découvrir les mots-clés relevant d'un thème "signal faible" éventuel. Pour atteindre cet objectif, nous proposons une définition du signal faible qui conditionne l'approche conjointe modèle thématique / plongement lexical, et contraint le choix des méthodes LDA et Word2Vec. Nous proposons d'évaluer les partitions obtenues grâce à un indice de cohérence sur la collection de mots représentative de chaque thème obtenu. Les clusters obtenus sont ainsi plus cohérents au sens contextuel. La détection du cluster associé au signal faible est alors plus aisée et plus pertinente.

International audience; Background: Amyotrophic Lateral Sclerosis (ALS) is a relentlessly progressive neurodegenerative condition with limited therapeutic options at present. Survival from symptom onset ranges from 3 to 5 years depending on genetic, demographic, and phenotypic factors. Despite tireless research efforts, the core etiology of the disease remains elusive and drug development efforts are confounded by the lack of accurate monitoring markers. Disease heterogeneity, late-stage recruitment into pharmaceutical trials, and inclusion of phenotypically admixed patient cohorts are some of the key barriers to successful clinical trials. Machine Learning (ML) models and large international data sets offer unprecedented opportunities to appraise candidate diagnostic, monitoring, and prognostic markers. Accurate patient stratification into well-defined prognostic categories is another aspiration of emerging classification and staging systems.Methods: The objective of this paper is the comprehensive, systematic, and critical review of ML initiatives in ALS to date and their potential in research, clinical, and pharmacological applications. The focus of this review is to provide a dual, clinical-mathematical perspective on recent advances and future directions of the field. Another objective of the paper is the frank discussion of the pitfalls and drawbacks of specific models, highlighting the shortcomings of existing studies and to provide methodological recommendations for future study designs.Results: Despite considerable sample size limitations, ML techniques have already been successfully applied to ALS data sets and a number of promising diagnosis models have been proposed. Prognostic models have been tested using core clinical variables, biological, and neuroimaging data. These models also offer patient stratification opportunities for future clinical trials. Despite the enormous potential of ML in ALS research, statistical assumptions are often violated, the choice of specific statistical models is seldom justified, and the constraints of ML models are rarely enunciated.Conclusions: From a mathematical perspective, the main barrier to the development of validated diagnostic, prognostic, and monitoring indicators stem from limited sample sizes. The combination of multiple clinical, biofluid, and imaging biomarkers is likely to increase the accuracy of mathematical modeling and contribute to optimized clinical trial designs.

Working paper Il s'agit d'un preprint qui a été soumis dans une revue à comité de lecture; This paper presents a novel algorithm for performing inference and/or clustering in semiparametric copula-based mixture models. The algorithm replaces the standard kernel density estimator by a weighted version that permits to take into account the constraints put on the underlying marginal densities. Lower misclassification error rates and better estimates are obtained on simulations. The pointwise consistency of the weighted kernel density estimator is established under an assumption on the rate of convergence of the sample maximum.

International audience; This paper is related to a project aiming at discovering weak signals from different streams of information, possibly sent by whistleblowers in a platform as GlobalLeaks. The study presented in this paper tackles the particular problem of clustering topics at multi-levels from multiple documents, and then extracting meaningful descriptors, such as weighted lists of words for document representations in a multi-dimensions space. In this context, we present a novel idea which combines Latent Dirichlet Allocation and Word2Vec (providing a consistency metric regarding the partitioned topics) as potential method for limiting the "a priori" number of cluster k usually needed in classical partitioning approaches. We proposed 2 implementations of this idea, respectively able to: (1) finding the best k for LDA in terms of topic consistency; (2) gathering the optimal clusters from different levels of clustering. We also proposed a non-traditional visualization approach based on a multi-agents system which combines both dimension reduction and interactivity.; L'étude présentée s'inscrit dans le cadre du développement d'une plateforme d'analyse automatique de documents associée à un service sécurisé lanceurs d'alerte, de type GlobalLeaks. Cet article se focalise principalement sur la recherche de signaux faibles présents dans les documents. Il s'agit d'une problématique investiguée dans un grand nombre de champs disciplinaires et de cadres applicatifs. Nous supposons que chaque document est un mélange d'un petit nombre de thèmes ou catégories, et que la création de chaque mot est attribuable en termes de probabilités à l'un des thèmes du document. Les catégories des documents transmis ne sont pas connues"a priori". Les mots-clés présents dans les documents représentatifs de ces catégories sont également inconnus. L'analyse des documents reçus doit simultanément permettre de découvrir les thèmes, classer les documents relativement à ces thèmes, détecter les mots-clés pertinents relatifs aux thèmes et enfin découvrir les mots-clés relevant d'un thème "signal faible" éventuel. Pour atteindre cet objectif, nous proposons une définition du signal faible qui conditionne l'approche conjointe modèle thématique / plongement lexical, et contraint le choix des méthodes LDA et Word2Vec. Nous proposons d'évaluer les partitions obtenues grâce à un indice de cohérence sur la collection de mots représentative de chaque thème obtenu. Les clusters obtenus sont ainsi plus cohérents au sens contextuel. La détection du cluster associé au signal faible est alors plus aisée et plus pertinente.

National audience; A new robust lane marking extraction algorithm for monocular vision is proposed based on Two-Dimension Declivity. It is designed for the urban roads with difficult conditions (shadow, high brightness, etc.). In this paper, we propose a locating system which, from an embedded camera, allows lateral positioning of a vehicle by detecting road markings. The primary contribution of the paper is that it supplies a robust method made up of six steps: (i) Image Pre-processing, (ii) Enhanced Declivity Operator (DE), (iii) Mathematical Morphology, (iv) Labeling, (v) Hough Transform and (vi) Line Segment Clustering. The experimental results have shown the high performance of our algorithm in various road scenes. This validation stage has been done with a sequence of simulated images. Results are very promising: more than 90% of marking lines are extracted for less than 12% of false alarm.

International audience; In this paper, we define a cluster based scheduling algorithm for Cognitive Radio Sensor Networks (CRSNs). To avoid inter-clusters collision, we assign fixed channels only to nodes having one-hop neighbors out of their clusters. We denote these nodes as specific nodes. Previous studies assign distinct channels to whole neighbor clusters to avoid inter-clusters collision. Our objective is to optimize the spatial reuse and to increase the network throughput while saving sensors energy. We start by assigning channels only to the specific nodes. Once the problem of inter-clusters collision is solved, each cluster head (CH) schedules the transmissions in its cluster independently. For the cluster members that are specific nodes, the CH assigns only time slots because the channel assignment is already done. For other cluster members (CMs) (not specific nodes), the CH assigns the pair (channel, slot). Two solutions are proposed in this paper to schedule the CMs: The Frame Intra Cluster Multichannel Scheduling algorithm denoted Frame-ICMS and the Slot Intra Cluster Multichannel Scheduling algorithm denoted Slot-ICMS. We evaluate the performance of these algorithms in case of accurate PUs activity detection and in case of bad PUs activity estimation. We prove that our proposals outperform an existing one especially in terms of energy saving.

International audience; Traditionally, clustering operates on data described by a fixed number of (usually numerical) features; this description schema is said propositional or attribute-value. Yet, when the data cannot be described in that way, usual data-mining or clustering algorithms are no longer suitable. In this paper, we consider the problem of discovering similar types of programs in TV streams. The TV data have two important characteristics: 1) they are multi-relational, that is to say with multiple relationships between features; 2) they require background knowledge external to their interpretation. To process the data, we use Inductive Logic Programming (ILP) [9]. In this paper, we show how to divert ILP to work unsupervised in this context: from artificial learning problems, we induce a notion of similarity between broadcasts, which is later used to perform the clustering. Experiments presented show the soundness of the approach, and thus open up many research avenues.

International audience; This paper deals with the problem of improving data-based diagnosis of continuous systems taking advantage of the system control information represented as discrete event dynamics. The approach starts from dynamic clustering results and, combining the information about operational modes, automatically generates a discrete event system that improves clustering results interpretability for decision-making purposes and enhances fault detection capabilities by the inclusion of event related dynamics. The generated timed discrete event system is adaptive thanks to the dynamic nature of the clusterer from which it was learned, namely DyClee. The timed discrete event system brings valuable temporal information to distinguish behaviors that are non-diagnosable based solely on the clustering itself.

International audience; This paper introduces a fast method to generate high performance parallelized code from a dataflow specification of an application. Dataflow Models of Computation (MoCs) are efficient programming paradigms for expressing the parallelism of an application. Traditionally, mapping and scheduling methods for dataflow MoCs rely on complex graph transformations to explicit their parallelism which can result in complex graph for embarrassingly parallel applications. For such applications, state-of-the-art mapping and scheduling techniques are prohibitively complex, while the exposed parallelism often exceeds the parallel processing capabilities of the target architecture. We propose SCAPE, an automated method to control the complexity of the pre-scheduling graph transformation by using information from the architecture and application models. By decreasing the complexity of the graph, the mapping scheduling task is accelerated at the potential expense of the produced schedule. Our method offers a limited and controlled decrease of the schedule quality while enabling mapping and scheduling execution time between 1 and 2 orders of magnitude faster than state-of-the-art techniques.; Les modèles flux de données sont des paradigmes de programmation efficaces pour exposer le parallélisme d’uneapplication. Traditionnellement, les méthodes de placement et d’ordonnancement pour les modèles flux de données reposent sur destransformations complexes de graphes pour exposer tout le parallélisme potentiel d’une application, ce qui peut donner lieu à desgraphes complexes pour des applications hautement parallèles. La complexité des algorithmes de placement et d’ordonnancementsur ces graphes complexes est prohibitive, alors que le parallélisme exposé dépasse souvent le parallélisme potentiel de l’architecturecible. Nous proposons dans cet article la méthode automatique SCAPE pour contrôler la complexité de la transformation du graphe depré-ordonnancement en utilisant des informations provenant des modèles d’architecture et d’application. En diminuant la complexitéde ce graphe, les algorithmes de placement et d’ordonnancement sont accélérés au détriment potentiel de l’ordonnancement produit.La méthode SCAPE permet de contrôler la taille du graphe de pré-ordonnancement pour trouver le compromis optimal entre rapiditéet qualité du placement-ordonnancement. La méthode SCAPE accélère les algorithmes de placement et de l’ordonnancement del’état de l’art d’un facteur pouvant aller jusqu’à 2 ordres de grandeur

International audience; This paper presents an original approach combining Artificial Neural Networks (ANNs) and clustering in order to detect pollutant peaks. We developed air quality forecasting models using machine learning methods applied to hourly concentrations of ozone (O 3), nitrogen dioxide (NO 2) and particulate matter (PM 10) 24 hours ahead. MultiLayer Perceptron (MLP) was used alone, then hybridized successively with hierarchical clustering and with a combination of self-organizing map and k-means clustering. Clustering methods were used to subdivide the dataset, and then an MLP was trained on each subset. Two urban sites of Corsica Island in the western Mediterranean Sea were investigated. These models showed a good global precision (Index of Agreement reaching 0.87 for O 3 , 0.80 for NO 2 and 0.74 for PM 10). Considering it is particularly important than forecasting model used on an operational basis correctly predict pollution peaks, a sensitivity analysis was performed using Receiver Operating Characteristic curves (ROC curves). It allowed to evaluate the behaviour and the robustness of the models for high concentration situations. The results show that for PM 10 and O 3 , hybrid models made of a combination of clustering and MLP outperform classical MLP most of the time for high concentration prediction. An operational tool has been built with the models presented in this paper, and is used for air quality forecasting in Corsica.

National audience; This paper takes a step towards the extraction of contrastive explanations between anoma- lies and the intrinsic structure of regular points. It proposes a variant of the isolation forest algorithm whose main objective is to preserve the structure of regular data in order to facilitate its reconstruction. Experiments conducted on synthetic datasets show that this variant of isola- tion forest deteriorates less the structure of regular data than the classical method. Therefore, the former can serve as a basis for a unified approach to anomaly detection and explanation.; Cet article effectue un pas vers une extraction d'explications contrastives entre anomalies et structure intrinsèque des points réguliers. Il propose une variante de l'algorithme des forêts d'isolation ayant pour objectif principal la préservation de la structure des données régulières en vue de sa reconstitution plus aisée. Les expérimentations menées sur des jeux de données synthétiques montrent que cette variante des forêts d'isolation détériore moins la structure des données régulières que la méthode classique. Par conséquent, la première citée peut servir de base pour une approche unifiée de détection et d'explication d'anomalies.

A large class of problems can be formulated as inverse problems, where the goal is to find parameter values that best explain some observed measures. Typical constraints in practice are that the relationships between parameters and observations are highly nonlinear, with high-dimensional observations and multi-dimensionally correlated parameters. To deal with these constraints, we consider probabilistic mixtures of locally linear models using inverse regression strategies, namely the Gaussian locally linear mapping (GLLiM) models. These can be seen as special instances of a mixture of experts (MoE) models. The popularity of MoE is largely due to their universal approximation properties, provided that the number of mixture components is large enough. In this paper, we propose a general scheme to design a tractable Bayesian nonparametric GLLiM (BNP-GLLiM) model to avoid any commitment to an arbitrary number of components. A tractable estimation algorithm is designed using a variational Bayesian expectation-maximization. In particular, we establish posterior consistency for the number of components in BNP-GLLiM after the merge-truncate-merge algorithm post-processing. Illustrations on simulated data show good results in terms of recovering the true number of clusters and the mean regression function.

International audience; In this paper, we use nonparametric runs-based tests to analyze the randomness and the persistence of relative returns of hedge funds. Runs tests are implemented on a universe of hedge funds extracted from HFR database over the period spanning January 2000 to December 2012. Our findings suggest that i) slightly less than 80% of the studied universe has returns at random, ii) a similar figure is found out when focusing on relative returns, iii) hedge funds that do present clustering in their relative returns are mainly found within Event Driven and Relative Value strategies, iv) and for relative returns, results vary with the type of the benchmark nature (peer group average or traditional). This paper also emphasizes that runs tests may be a useful tool for investors in their fund's selection process.

TRB - Transportation Research Board 94th Annual Meeting, WASHINGTON DC, ETATS-UNIS, 11-/01/2015 - 15/01/2015; Determining travel times on arterials is crucial for urban traffic management. Travel times strongly depend on traffic conditions and tend to highly increase when congestion appears. Most existing works consider congested situations that occur due to peaks of the upstream demand and not to downstream spillbacks. Only the second option leads to fully congested situations on the whole arterial. This paper explains that considering imaginary voids instead of vehicles is more natural in congestion. Solving the problem becomes then easy. Voids travel toward update stream. They move at the backward wave speed when passing each other a traffic flow at capacity, stopping in absence of vehicle, and traveling at a reduced speed when encountering a generic traffic flow. Using this concept, an existing method of travel time computation is extended to congestion. In free-flow conditions, this method consists in determining the separation between vehicle groups at each signal of the arterial. The main point to apply this method in congestion is to consider the voids instead of vehicles. Besides, this paper generalizes the Flow-Platoon Diagram (FPD) concept introduced in a previous work. It aggregates all possible travel times of vehicles in all possible stationary conditions on an arterial. Moreover, it easily leads to an exact estimation of the Macroscopic Fundamental Diagram. Input data needed to use the proposed approach are fundamental diagram and signal timings. After that, FPD provides travel times according to the flow from loop detectors.

Energy optimization has become a crucial issue in the realm of ICT. This paper addresses the problem of energy consumption in a Metro Ethernet network. Ethernet technology deployments have been increasing tremendously because of their simplicity and low cost. However, much research remains to be conducted to address energy efficiency in Ethernet networks. In this paper, we propose a novel Energy Aware Forwarding Strategy for Metro Ethernet networks based on a modification of the Internet Energy Aware Routing (EAR) algorithm. Our contribution identifies the set of links to turn off and maintain links with minimum energy impact on the active state. Our proposed algorithm could be a superior choice for use in networks with low saturation, as it involves a tradeoff between maintaining good network performance and minimizing the active links in the network. Performance evaluation shows that, at medium load traffic, energy savings of 60% can be achieved. At high loads, energy savings of 40% can be achieved without affecting the network performance.

6 pages; International audience; Traditional Network Intrusion Detection Systems (NIDSs) rely on either specialized signatures of previously seen attacks, or on expensive and difficult to produce labeled traffic datasets for profiling and training. Both approaches share a common downside: they require the knowledge provided by an external agent, either in terms of signatures or as normal-operation profiles. In this paper we describe UNIDS, an Unsupervised NIDS capable of detecting 0-day attacks, i.e., network attacks for which no signature is yet available, without using any kind of signatures, labeled traffic, or training. UNIDS uses a novel unsupervised outliers detection approach based on Sub-Space Clustering and Multiple Evidence Accumulation techniques to pin-point different kinds of network intrusions and attacks such as DoS/DDoS, probing attacks, propagation of worms, buffer overflows, illegal access to network resources, etc. In this paper we make the strong point that the de-facto approach for NIDS, namely the application of rule-based detection techniques, can be highly harmful for the protected network in case of 0-day attacks. In contrast, we show how UNIDS can work as a complementary system to current NIDS to detect the occurrence of previously unseen attacks. For doing so, we compare the performance of a standard rule-based NIDS against UNIDS to detect 0-day attacks in the well-known KDD99 dataset. In addition, we also compare the performance of UNIDS against other popular unsupervised detection techniques to detect attacks in traces collected at two operation networks.

International audience; Pervasive Information System (PIS) represents a new generation of Information Systems (IS) available anytime, anywhere in a pervasive environment. In this paper, we propose to enhance PIS transparency and efficiency through a context-aware intentional service prediction approach. This approach allows anticipating user's future needs, offering and recommending him the most suitable service in a transparent and discrete way. We detail in this paper our service prediction mechanism and present encouraging experimental results demonstrating our proposition.

International audience; Wireless sensors networks (WSNs) are traditionally composed of large number of low-cost and tiny homogenous sensors nodes connected through a wireless network that gather data to be treated locally or relayed to the sink node through multi-hop wireless transmission. Moreover, such issues are very critical due servers resources constraints like efficient energy, stock limitation and lifetime of network. Several solutions were proposed to minimize the traffic into network. Clustering algorithms have been widely used to reduce energy consumption. In this context, the key point in such topology is to select a cluster. One of solutions is to select a cluster alternately. However, this choice does not consider the energy as important criteria in actual papers. In order to limit energy consumption, our new method is proposed in this paper to optimization Low Energy Adaptive Clustering Hierarchy (O- LEACH) to improve existing LEACH and LEACH-C by selecting cluster according to the residual energy of nodes dynamically. The simulation results show that proposed algorithm achieve longer stability by comparison to original LEACH and LEACH-C.

TRB 2014 - 93rd Transportation Research Board Annual Meeting, WASHINGTON DC, ETATS-UNIS, 12-/01/2014 - 16/01/2014; The travel time variability may significantly influence the driver's route choice. Estimating travel time reliability is thus really appealing to urban infrastructure managers. This could provide better information to drivers, encourage their reassignment on the network, and optimize its use. In this paper, we study any arterial in under-saturated conditions with known fixed-time traffic signals. The kinematic wave model is chosen to simply represent traffic dynamics. A semi-analytical (grid-free) method is applied to determine the separation between vehicle groups at each signal of the arterial. The main contribution of this paper is to present an aggregated diagram that describes the functioning of this arterial. It graphically provides the direct assessment of vehicle travel times with respect to their departure and the traffic flow. This tool only depends on signal settings and is cyclic and invariant for fixed-time signals. We show how to use this diagram to generate probabilistic travel time distributions when some input parameters are uncertain. Furthermore, this diagram appears insightful when the traffic flow is not accurately known.

International audience; Leachate recirculation is a key process in municipal waste landfills functioning as bioreactors. To quantify the water content and to assess the leachate injection system, in-situ methods are required to obtain spatially distributed information, usually electrical resistivity tomography (ERT). This geophysical method is based on the inversion process, which presents two major problems in terms of delimiting the infiltration area. First, it is difficult for ERT users to choose an appropriate inversion parameter set. Indeed, it might not be sufficient to interpret only the optimum model (i.e. the model with the chosen regularisation strength) because it is not necessarily the model which best represents the physical process studied. Second, it is difficult to delineate the infiltration front based on resistivity models because of the smoothness of the inversion results. This paper proposes a new methodology called MICS (multiple inversions and clustering strategy), which allows ERT users to improve the delimitation of the infiltration area in leachate injection monitoring. The MICS methodology is based on (i) a multiple inversion step by varying the inversion parameter values to take a wide range of resistivity models into account and (ii) a clustering strategy to improve the delineation of the infiltration front. In this paper, MICS was assessed on two types of data. First, a numerical assessment allows us to optimise and test MICS for different infiltration area sizes, contrasts and shapes. Second, MICS was applied to a field data set gathered during leachate recirculation on a bioreactor.

International audience; K-medoids clustering is solvable by dynamic programming in O(N 3) time for a 2D Pareto Front (PF). A key element is a interval clustering optimality. This paper proves this property holds also for local minima for k-medoids. It allows to enumerate the local minima of k-medoids with the same complexity than the computation of global optima for k=2 ou k=3. A pseudo-polynomial enumeration scheme is designed, for small values of k. This allows to understand results obtained by local search approaches in a 2D PF.

The problem of adaptive noisy clustering is investigated. Given a set of noisy observations $Z_i=X_i+\epsilon_i$, $i=1,...,n$, the goal is to design clusters associated with the law of $X_i$'s, with unknown density $f$ with respect to the Lebesgue measure. Since we observe a corrupted sample, a direct approach as the popular {\it $k$-means} is not suitable in this case. In this paper, we propose a noisy $k$-means minimization, which is based on the $k$-means loss function and a deconvolution estimator of the density $f$. In particular, this approach suffers from the dependence on a bandwidth involved in the deconvolution kernel. Fast rates of convergence for the excess risk are proposed for a particular choice of the bandwidth, which depends on the smoothness of the density $f$. Then, we turn out into the main issue of the paper: the data-driven choice of the bandwidth. We state an adaptive upper bound for a new selection rule, called ERC (Empirical Risk Comparison). This selection rule is based on the Lepski's principle, where empirical risks associated with different bandwidths are compared. Finally, we illustrate that this adaptive rule can be used in many statistical problems of $M$-estimation where the empirical risk depends on a nuisance parameter., Comment: 22 pages

The disambiguation of named entities is a challenge in many fields such as scientometrics, social networks, record linkage, citation analysis, semantic web...etc. The names ambiguities can arise from misspelling, typographical or OCR mistakes, abbreviations, omissions...So the search of names of persons or of organization is difficult, a single name can appear in different forms. This paper proposes two approaches to disambiguate on the affiliations of authors of scientific papers in bibliographic databases: the first way, considers that we have a training corpus, and uses a Naive Bayesian model. The second way assumes that we have not resource learning, and uses a semisupervised approach, mixing soft-clustering and Bayesian learning. The results are encouraging and are already partially applied in a scientific survey department. However, we aware that our approach may have limitations: we can't process efficiently highly unbalanced data but solutions are possible for future developments., La désambiguïsation d'entités nommées est un défi dans de nombreux domaines tels que la scientométrie, les réseaux sociaux, l'analyse des citations, le Web sémantique etc ... Les ambiguïtés peuvent provenir de fautes d'orthographe, d’erreurs typographiques ou d’OCR, d’abréviations, d’omissions ... Ainsi, la recherche de noms de personnes ou d’organisations est rendue difficile par la multiplicité des formes utilisées. Cet article propose deux approches pour lever l'ambiguïté sur les affiliations des auteurs d'articles scientifiques dans des bases de données bibliographiques : la première, considère un corpus d'apprentissage, et utilise un modèle bayésien naïf ; la deuxième sans ressource d'apprentissage, une approche semi-supervisée, associant une méthode de clustering recouvrant et un apprentissage bayésien. Les résultats sont encourageants et les méthodes sont déjà partiellement appliquées dans un pôle de veille scientifique. Cependant, cette approche a des limites : par exemple, on ne peut pas traiter efficacement des données très déséquilibrées, mais des solutions sont envisageables pour de futurs développements.

International audience; The disambiguation of named entities is a challenge in many fields such as scientometrics, social networks, record linkage, citation analysis, semantic web...etc. The names ambiguities can arise from misspelling, typographical or OCR mistakes, abbreviations, omissions... Therefore, the search of names of persons or of organizations is difficult as soon as a single name might appear in many different forms. This paper proposes two approaches to disambiguate on the affiliations of authors of scientific papers in bibliographic databases: the first way considers that a training dataset is available, and uses a Naive Bayes model. The second way assumes that there is no learning resource, and uses a semi-supervised approach, mixing soft-clustering and Bayesian learning. The results are encouraging and the approach is already partially applied in a scientific survey department. However, our experiments also highlight that our approach has some limitations: it cannot process efficiently highly unbalanced data. Alternatives solutions are possible for future developments, particularly with the use of a recent clustering algorithm relying on feature maximization.

International audience; The disambiguation of named entities is a challenge in many elds such as sciento- metrics, social networks, record linkage, citation analysis, semantic web...etc. The names ambiguities can arise from misspelling, typographical or OCR mistakes, abbreviations, omissions... So the search of names of persons or of organization is di cult, a single name can appear in di erent forms. This paper proposes two approaches to disambiguate on the a liations of authors of sci- enti c papers in bibliographic databases: the rst way, considers that we have a training corpus, and uses a Naive Bayesian model. The second way assumes that we have not re- source learning, and uses a semi-supervised approach, mixing soft-clustering and Bayesian learning. The results are encouraging and are already partially applied in a scienti c survey department. However, we aware that our approach may have limitations: we can't process e ciently highly unbalanced data but solutions are possible for future developments.

International audience; Collaborative innovation is an unavoidable need for the small and medium enterprises (SME) both in terms of economic scale and technological knowledge. Risks and the innovation power are analyzed for the wealth of collaboration. This paper aims to present the synergy index as a multiplier of the innovation power of research partners to construct a successful collaboration. The proposed index can be used with different number of companies in collaboration cluster and the synergy maximization is guaranteed by using a new particle swarm algorithm, Foraging Search. This paper will give the formulation and criteria of the synergy index in detail. A sample synergy index application for the Turkish SMEs will clarify the steps to follow.

3 pages, presented at the 1st International Workshop for Languages for Distributed Algorithms (LADA-2012), Philadelphia, USA, 23-24 January 2012, co-located with POPL 2012: 39th ACM SIGPLAN-SIGACT Symp. on Principles of Programming Languages; The growing limitations of traditional social networks have prompted work on novel and fully decentralised alternatives to implement on-line social applications. To realise these decentralised social networks, gossip protocols appear as a natural solution, as they intrinsically tend to be highly resilient, efficient, and scalable. Existing gossip-based social networks have however so far been limited to simple use-cases, that typically take a uniform view of users, peers, and the data they hold. To progress to full-fledged decentralised social networks, we posit in this paper that we now need to move to gossip-based social applications that can simultaneously cater for different types of data and services. To help in this task, this paper sketches the main ingredients of a new programming language, Constellation, that seeks to simplify the realisation and experimentation with modular social gossip-based applications.

This paper focuses on unsupervised curve classification in the context of nuclear industry. At the Commissariat à l'Energie Atomique (CEA), Cadarache (France), the thermal-hydraulic computer code CATHARE is used to study the reliability of reactor vessels. The code inputs are physical parameters and the outputs are time evolution curves of a few other physical quantities. As the CATHARE code is quite complex and CPU-time consuming, it has to be approximated by a regression model. This regression process involves a clustering step. In the present paper, CATHARE output curves are clustered using a k-means scheme, with a projection onto a lower dimensional space. We study the properties of the empirically optimal cluster centers found by the clustering method based on projections, compared to the “true” ones. The choice of the projection basis is discussed, and an algorithm is implemented to select the best projection basis among a library of orthonormal bases. The approach is illustrated on a simulated example and then applied to the industrial problem.

International audience; This paper proposes a new iterative algorithm for Monte Carlo criticality calculations based on the Adaptative Multilevel Splitting (AMS) algorithm initially developed as a variance reduction technique. The objective is to modify the AMS and use it in place of the standard Power Iteration technique, aiming at reducing the variance, the bias, the spatial and cycle-to-cycle correlations. As a first step towards proof of concept, the method has been tested on one-dimensional homogeneous rods of different sizes, in a mono-energetic framework. The preliminary results presented here suggest that the method could provide promising improvements regarding cycle-to-cycle and spatial correlations, and thus reducing clustering. The AMS combined with branchless collision allowed for an increase in the Figure of Merit by a factor 63 for the keff, and by three order of magnitude on average for the spatial flux (from 4 to 105) on a 80 cm rod.

National audience; This paper describes a novel technique to cluster Out-Of-Vocabulary (OOV) word tokens in a LVCSR system used for transcribing broadcast news speech data. The system is composed of two blocks: (1) an OOV word detector and (2) a clustering module working on the detected OOV word segments. This combination allows a more reliable detection of repeated OOV words than would be possible with the OOV detector only. In the paper we focus our attention on the second part of the system i.e. the cluster-ing algorithm. This algorithm is based on the estimation of the entropy. The proposed algorithm gives better per-formance than a classical incremental clustering algorithm based on a distance threshold.

ISSN: 1828-6003, CD ISSN: 1828-6011; International audience; The topic of research exposed in this paper concerns Content Based image retrieval in a heterogeneous high database. The increase of storage capacities and the evolution of compression techniques have generated an explosion of the digital information quantity. Their computerization opens a vast field of applications. In this setting we are interested more especially in the problem of the dimensionality reduction and spectral clustering of a heterogeneous database of images in order to image retrieval by the content. Our new gait described in this paper consists to: in first phase the description of the database images by a hybrid descriptor which are Interest SIFT points combined with texture descriptor given by the application of Wavelet transform. The descriptor is multi-dimensional, robust and invariant to changes and scales. In second phase the representation of the database images as a convex graph. In third phase the reduction of the space of representation by the application of an unsupervised spectral classification (The Spectral training uses information contained in the eigenvectors of the normalized matrix of transition to detect structures in data.) That will provide us classes of images that has shortcoming the Eigen-values calculated on the matrix of symmetry. As last phase, we use the Nyström theory that will permit us, not to recalculate the all Eigen-values, but only the lasts one.

International audience; We propose, in this paper, a new biometric identification approach which aims to improve recognition performances in identification systems. We aim to split the identity database into well separated partitions in order to simplify the identification task. In this paper we develop a face identification system and we use the reference algorithms of eigenfaces and fisherfaces in order to extract different features describing each identity. These features, which describe faces, are generally optimized to establish the required identity in a classical identification process. In this work, we develop a novel criterion to extract features used to partition the identity database. We develop database partitioning with clustering methods which split the gallery by bringing together identities which have similar features and separating dissimilar features in different bins. Pruning the most dissimilar bins from the query identity features allows us to improve the identification performances. We report results from the XM2VTS database.