Publications

2017

• Semi-Distributed Demand Response Solutions for Smart Homes
  
  • Kaddah Rim
  • Kofman Daniel
  • Mathieu Fabien
  • Pioro Michal
  , 2017, pp.17 (163--179). The Internet of Things (IoT) paradigm brings an opportunity for advanced Demand Response (DR) solutions. It enables visibility and control on the various appliances that may consume, store or generate energy within a home. It has been shown that a centralized control on the appliances of a set of households leads to efficient DR mechanisms; unfortunately, such solutions raise privacy and scalability issues. In this chapter we propose an approach that deals with these issues. Specifically, we introduce a scalable two-levels control system where a centralized controller allocates power to each house on one side and, each household implements a DR local solution on the other side. A limited feedback to the centralized controller allows to enhance the performance with little impact on privacy. The solution is proposed for the general framework of capacity markets. (10.1007/978-981-10-1741-4_12)
  DOI : 10.1007/978-981-10-1741-4_12
• Two-Stage volumetric texture synthesis based on structural information
  
  • Akl Adib
  • Yaacoub Charles
  • Donias Marc
  • da Costa Jean-Pierre
  • Germain Christian
  , 2017, pp.1-6. (10.1109/IPTA.2017.8310141)
  DOI : 10.1109/IPTA.2017.8310141
• Nouveau récepteur radio numérique pour les observations astrophysiques spatiales dans la bande de fréquence 1 kHz à 50 MHz
  
  • Gargouri Yosra
  , 2017. Plusieurs phénomènes astronomiques émettent des ondes radios basses fréquences tels que les éruptions solaires, les magnétosphères, les pulsars . . . Certains de ces ondes sont mal captées par les observatoires terrestres à cause, principalement, de la coupure ionosphérique. Il devient indispensable d’envoyer des récepteurs radio dans l’espace pour les acquérir. Cependant, ces récepteurs sont consommation d’énergie et le taux de transmission. Un paradigme récent pour l’acquisition et la reconstruction des signaux, appelé l’échantillonnage comprimé (Compressive sampling, Compressed Sensing, CS) pourra être une réponse adéquate à ces problématiques en limitant, dès l’acquisition, la quantité de données numérisés : En effet, le CS a permis l’émergence d’un nouveau type de Convertisseur Analogique-Numérique (ADC) appelé Convertisseur Analogique-Information (AIC) qui permet d’échantillonner à une fréquence potentiellement inférieure à celle prescrite par Nyquist-Shannon, en exploitant le caractère parcimonieux des signaux. Nous proposons dans le cadre de cette thèse d’étudier l’application de l’échantillonnage comprimé pour l’acquisition des signaux astrophysiques spatiaux dans la bande de fréquence [1kHz à 50 MHz]. Nous nous focalisons sur des signaux émis par les deux sources radio les plus brillantes dans le ciel telles que vues de la Terre, à savoir le Soleil et Jupiter. En se basant sur les propriétés caractéristiques de nos signaux d’intérêt, nous avons construit progressivement et méthodologiquement notre schéma d’acquisition : En commençant par l’étude de compressibilité des signaux, puis l’identification de l’architecture du Convertisseur Analogique-Information (AIC) appropriée et enfin le choix de l’algorithme de reconstruction du signal. Nous avons également proposé une nouvelle implémentation flexible et programmable de l’AIC retenu, qui permet l’acquisition de différents types de signal ayant le même domaine de compressibilité avec différents facteurs de compression. En utilisant une technologie CMOS 65 nm, nous avons évalué le gain en quantité de données acquise et en consommation de puissance de cette architecture par rapport au convertisseur analogique-numérique traditionnel.
• TLS for Cooperative ITS Services
  
  • Msahli Mounira
  • Serhrouchni Ahmed
  • Labiod Houda
  • Kaiser Arnaud
  • Lonc Brigitte
  , 2017, pp.176-189. (10.1007/978-3-319-72329-7_16)
  DOI : 10.1007/978-3-319-72329-7_16
• Nonnegative Feature Learning Methods for Acoustic Scene Classification
  
  • Bisot Victor
  • Serizel Romain
  • Essid Slim
  • Richard Gael
  , 2017. This paper introduces improvements to nonnegative feature learning-based methods for acoustic scene classification. We start by introducing modifications to the task-driven nonnegative matrix factorization algorithm. The proposed adapted scaling algorithm improves the generalization capability of task-driven nonneg-ative matrix factorization for the task. We then propose to exploit simple deep neural network architecture to classify both low level time-frequency representations and unsupervised nonnegative matrix factorization activation features independently. Moreover, we also propose a deep neural network architecture that exploits jointly unsupervised nonnegative matrix factorization activation features and low-level time frequency representations as inputs. Finally, we present a fusion of proposed systems in order to further improve performance. The resulting systems are our submission for the task 1 of the DCASE 2017 challenge.
• Modèles Markoviens pour les images SAR : application à la détection de l'eau dans les images satellitaires SWOT et analyse multi-temporelle de zones urbaines
  
  • Lobry Sylvain
  , 2017. Afin d’obtenir une meilleure couverture, à la fois spatiale et temporelle de leurs mesures les hydrologues utilisent des données spatiales en plus de celles acquises sur place. Fruit d’une collaboration entre les agences spatiales française (le CNES) et américaine (JPL, NASA), la future mission SWOT a notamment pour but de fournir des mesures de hauteur des surfaces d’eau continentales en utilisant l’interférométrie radar à synthèse d’ouverture (SAR). Dans cette thèse, nous nous intéressons au problème de la détection de l’eau dans les images d’amplitude SWOT qui est ici un prérequis au traitement interférométrique. Dans cette optique, nous proposons d’utiliser une méthode dédiée à la détection des larges cours d’eau ainsi qu’un traitement spécifique pour la détection de rivières fines. La première méthode est basée sur un champ de Markov (MRF) pour la classification, conjointement à une estimation des paramètres de classes qui ne peuvent être supposés constants dans le cas de SWOT. L’estimation des paramètres peut également être modélisée par des champs de Markov. La seconde méthode s’appuie sur une détection de segments au niveau pixellique complétée par une connexion de ces segments. Afin d’étudier l’extension aux séries multi-temporelles, nous proposons des méthodes de traitement adaptées aux données SAR de zones urbaines. Ces zones présentent de forts rétro-diffuseurs, ayant une radiométrie largement supérieure à celle des autres points dans l’image. Les modèles présentés prennent explicitement en compte la présence de ces forts rétro-diffuseurs en considérant les images comme une somme de deux composantes (le fond et les cibles fortes). Différents termes de régularisation peuvent alors être utilisés pour chacune de ces deux composantes. Modélisés comme des champs de Markov, ils peuvent alors être optimisés exactement par recherche de coupure minimale dans un graphe. Nous présentons des applications en détection de cibles fortes, régularisation et détection de changement dans ces séries.
• Data sparse nonparametric regression with epsilon-insensitive losses
  
  • Sangnier Maxime
  • Fercoq Olivier
  • d'Alché-Buc Florence
  , 2017, 77, pp.192-207. Leveraging the celebrated support vector regression (SVR) method, we propose a unifying framework in order to deliver regression machines in reproducing kernel Hilbert spaces (RKHSs) with data sparsity. The central point is a new definition of epsilon-insensitivity, valid for many regression losses (including quantile and expectile regression) and their multivariate extensions. We show that the dual optimization problem to empirical risk minimization with epsilon-insensitivity involves a data sparse regularization. We also provide an analysis of the excess of risk as well as a randomized coordinate descent algorithm for solving the dual. Numerical experiments validate our approach.
• One Class Splitting Criteria for Random Forests
  
  • Goix Nicolas
  • Drougard Nicolas
  • Brault Romain
  • Chiapino Mael
  , 2017, pp.pp. 1-16. Random Forests (RFs) are strong machine learning tools for classification and regression. However, they remain supervised algorithms, and no extension of RFs to the one-class setting has been proposed, except for techniques based on second-class sampling. This work fills this gap by proposing a natural methodology to extend standard splitting criteria to the one-class setting, structurally generalizing RFs to one-class classification. An extensive benchmark of seven state-of-the-art anomaly detection algorithms is also presented. This empirically demonstrates the relevance of our approach.
• Cooperative Communications in very large cellular Networks
  
  • David Alvarez-Corrales Luis
  , 2017. Recent studies have set the problem of base station cooperation within the framework of stochastic geometry, where the irregularity of the base station positions can be considered. Some authors study the case when the user can dynamically choose the set of stations cooperating for its service. This assumption is not realistic. Instead, other authors propose to form the groups in a static way. To be optimal, these static methodologies should consider proximity between the base stations to form the groups. We propose a grouping method based on the nearest neighbor model. We allow the formation of singles and pairs of nodes. We derive structural characteristics for these two processes and analyse the resulting interference fields. When the node positions are modelled by a Poisson point process, the processes of singles and pairs are not Poisson, complicating the corresponding analysis. The performance of the original model, however, can be approximated by the superposition of two Poisson point processes. Numerical evaluation shows coverage gains from different signal cooperation that can reach up to $15\%$, compared with the standard noncooperative case. For the cooperation to be meaningful, each station in a group should have sufficient resources to share, besides being close to each other. Thus, we redefine the nearest neighbors with a metric. The results of our analysis illustrate that cooperation gains strongly depend on the distribution of the available resources over the network.
• Cooperative communications in very large cellular networks
  
  • Alvarez Corrales Luis
  , 2017. Recent studies have set the problem of base station cooperation within the framework of stochastic geometry, where the irregularity of the base station positions can be considered. Some authors study the case when the user can dynamically choose the set of stations cooperating for its service. This assumption is not realistic. Instead, other authors propose to form the groups in a static way. To be optimal, these static methodologies should consider proximity between the base stations to form the groups. We propose a grouping method based on the nearest neighbor model. We allow the formation of singles and pairs of nodes. We derive structural characteristics for these two processes and analyse the resulting interference fields. When the node positions are modelled by a Poisson point process, the processes of singles and pairs are not Poisson, complicating the corresponding analysis. The performance of the original model, however, can be approximated by the superposition of two Poisson point processes. Numerical evaluation shows coverage gains from different signal cooperation that can reach up to 15%, compared with the standard noncooperative case. For the cooperation to be meaningful, each station in a group should have sufficient resources to share, besides being close to each other. Thus, we redefine the nearest neighbors with a metric. The results of our analysis illustrate that cooperation gains strongly depend on the distribution of the available resources over the network.
• Energy minimization based resource scheduling for strict delay constrained wireless communications
  
  • Fawaz Ibrahim
  • Ciblat Philippe
  • Sarkiss Mireille
  , 2017. This paper investigates the energy consumption minimization for resource scheduling in a wireless communication. We propose to take into account a strict delay constraint for each queued packet rather than an average delay constraint, in addition to a buffer overflow constraint. The associated optimization problem can be modeled as Constraint Markov Decision Problem where the actions are the number of packets sent on the known channel at each slot. The optimal random policy is exhibited through the resolution of standard linear programming. We show the gain in energy is substantial compared to naive policy. (10.1109/GlobalSIP.2017.8308645)
  DOI : 10.1109/GlobalSIP.2017.8308645
• Multiple Local Community Detection
  
  • Hollocou Alexandre
  • Bonald Thomas
  • Lelarge Marc
  , 2017. Community detection is a classical problem in the field of graph mining. We are interested in local community detection where the objective is the recover the communities containing some given set of nodes, called the seed set. While existing approaches typically recover only one community around the seed set, most nodes belong to multiple communities in practice. In this paper, we introduce a new algorithm for detecting multiple local communities, possibly overlapping, by expanding the initial seed set. The new nodes are selected by some local clustering of the graph embedded in a vector space of low dimension. We validate our approach on real graphs, and show that it provides more information than existing algorithms to recover the complex graph structure that appears locally.
• UE-HRI: a new dataset for the study of user engagement in spontaneous human-robot interactions
  
  • Ben-Youssef Atef
  • Clavel Chloé
  • Essid Slim
  • Bilac Miriam
  • Chamoux Marine
  • Lim Angelica
  , 2017, pp.464-472. (10.1145/3136755.3136814)
  DOI : 10.1145/3136755.3136814
• ALGeoSPF: A Hierarchical Geographical Factorization Model for POI Recommendation
  
  • Naacke Hubert
  • Griesner Jean-Benoît
  • Abdessalem Talel
  • Dosne Pierre
  , 2017, pp.11.
• Power excursion reduction in Flex-Grid optical networks with symbol rate adaptation
  
  • Amar Djamel
  • Samadi Payman
  • Bergman Keren
  • Lepers Catherine
  • Ware Cédric
  • Gravey Philippe
  , 2017. In this work, we propose a new use-case of symbol rate adaptation in Flex-Grid optical networks. We demonstrate that symbol rate adaptation of variable rate transponders can mitigate optical power excursion after spectrum defragmentation. (10.1364/ACPC.2017.S4C.5)
  DOI : 10.1364/ACPC.2017.S4C.5
• An Adaptive Distributed Asynchronous Algorithm with Application to Target Localization
  
  • Mourya Rahul
  • Bianchi Pascal
  • Salim Adil
  • Richard Cédric
  , 2017. This paper introduces a constant step size adaptive algorithm for distributed optimization on a graph. The algorithm is of diffusion-adaptation type and is asynchronous: at every iteration , some randomly selected nodes compute some local variable by means of a proximity operator involving a locally observed random variable, and share these variable with neighbors. The algorithm is built upon a stochastic version of the Douglas-Rachford algorithm. A practical application to target localization using measurements from multistatic continuous active sonar systems is investigated at length. (10.1109/CAMSAP.2017.8313117)
  DOI : 10.1109/CAMSAP.2017.8313117
• Optimal Transport to Rényi Entropies
  
  • Rioul Olivier
  , 2017, 10589. (10.1007/978-3-319-68445-1_17)
  DOI : 10.1007/978-3-319-68445-1_17
• Hypothesis Testing over Cascade Channels
  
  • Salehkalaibar Sadaf
  • Wigger Michèle
  • Wang Ligong
  , 2017, pp.369-373. Binary hypothesis testing over single and parallel cascade channels is considered where sensors communicate with dedicated relays, and these relays with a single final receiver. All relays as well as the final receiver decide on the binary hypothesis governing the joint probability distribution of the observations at the sensors, relays, and final receiver. The quantity of interest is the set of feasible type-II error exponents that allow for the type-I error probabilities to vanish asymptotically as the observation length increases. A coding scheme is proposed and the corresponding set of feasible type-II error exponents is analyzed by means of a modified Han-type analysis that can account for distributed decisions based on different codebooks and for nodes forwarding their decisions to other nodes. The obtained exponent region is optimal in some special cases. (10.1109/itw.2017.8277994)
  DOI : 10.1109/itw.2017.8277994
• Coded caching for wiretap broadcast channels
  
  • Kamel Sarah
  • Wigger Michèle
  • Sarkiss Mireille
  , 2017. The paper studies the wiretap erasure broadcast channel (BC) with an external eavesdropper when the legitimate receivers have cache memories. Various secure coding schemes are proposed for a scenario where Kw weak receivers have same erasure probabilities and Ks strong receivers have same erasure probabilities. The coding schemes achieve the cache-aided secrecy capacity when only weak receivers have cache memories and this cache memory is either small or large. They also allow to conclude the following: 1) Under a total cache budget it is often beneficial to assign the cache memories unequally between strong and weak receivers. 2.) Joint cache-channel coding is necessary to attain the optimal performance. 3.) The secrecy capacity can be positive even when the eavesdropper is stronger than the legitimate receivers. (10.1109/ITW.2017.8278037)
  DOI : 10.1109/ITW.2017.8278037
• Asymptotic High-SNR Capacity of MISO Optical Intensity Channels
  
  • Moser Stefan M
  • Wang Ligong
  • Wigger Michèle
  , 2017. This paper derives the asymptotic capacity for the multiple-input single-output free-space optical intensity channel in the regime of high signal-to-noise ratio (SNR). The asymptotic result is proven via upper and lower bounds on capacity at finite SNR. (10.1109/ITW.2017.8277933)
  DOI : 10.1109/ITW.2017.8277933
• Decentralized Access Control Mechanism with Temporal Dimension Based on Blockchain
  
  • Jemel Mayssa
  • Serhrouchni Ahmed
  , 2017, pp.177-182. Used mainly for the virtual money, the Blockchain technology adopts a decentralized network of peers to ensure a secure and transparent information storage and transmission. The basic use of Blockchain can be bypassed, and it is interesting to integrate it into other fields such as the Cloud storage. Cloud storage solutions ensure a continuous data synchronization and guarantee data sharing between different users. However, sharing user side encrypted data raises key and access control management challenges. Within this paper, we propose a novel access control model called Timely CP-ABE. Two main features come with our model. First, we introduce a decentralized access control mechanism where the user legitimacy is verified by Blockchain nodes. Second, we add temporal dimension to file sharing based on CP-ABE. In fact, we introduce a validity time to the access authorization without additional revocation cost. As a proof of concept, the implementation of the Timely CP-ABE based on Blockchain is performed on the CP-ABE toolkit and Multichain solution. (10.1109/ICEBE.2017.35)
  DOI : 10.1109/ICEBE.2017.35
• Decentralized Frank–Wolfe Algorithm for Convex and Nonconvex Problems
  
  • Wai Hoi-To
  • Lafond Jean
  • Scaglione Anna
  • Moulines Éric
  IEEE Transactions on Automatic Control, Institute of Electrical and Electronics Engineers, 2017, 62 (11), pp.5522 - 5537. Decentralized optimization algorithms have received much attention due to the recent advances in network information processing. However, conventional decentralized algorithms based on projected gradient descent are incapable of handling high-dimensional constrained problems, as the projection step becomes computationally prohibitive. To address this problem, this paper adopts a projection-free optimization approach, a.k.a. the Frank-Wolfe (FW) or conditional gradient algorithm. We first develop a decentralized FW (DeFW) algorithm from the classical FW algorithm. The convergence of the proposed algorithm is studied by viewing the decentralized algorithm as an inexact FW algorithm. Using a diminishing step size rule and letting t be the iteration number, we show that the DeFW algorithm's convergence rate is O(1/t) for convex objectives; is O(1/t2) for strongly convex objectives with the optimal solution in the interior of the constraint set; and is O(1/√t) toward a stationary point for smooth but nonconvex objectives. We then show that a consensus-based DeFW algorithm meets the above guarantees with two communication rounds per iteration. We demonstrate the advantages of the proposed DeFW algorithm on low-complexity robust matrix completion and communication efficient sparse learning. Numerical results on synthetic and real data are presented to support our findings. (10.1109/TAC.2017.2685559)
  DOI : 10.1109/TAC.2017.2685559
• Performance Study of View Synthesis with Small Baseline for Free Navigation
  
  • Nikitin Pavel
  • Jung Joel
  • Cagnazzo Marco
  • Pesquet-Popescu Beatrice
  , 2017.
• Passive Chaos Bandwidth Enhancement Under Dual-Optical Feedback with Hybrid III–V/Si DFB Laser
  
  • Schires Kevin
  • Gomez Sandra
  • Gallet Antonin E
  • Duan Guang-Hua
  • Grillot Frederic
  IEEE Journal of Selected Topics in Quantum Electronics, Institute of Electrical and Electronics Engineers, 2017, 23 (6), pp.1801309. The chaotic dynamics of a DFB laser are studied experimentally under a combination of short and long feedbacks. Chaos bandwidth enhancement is demonstrated using a hybrid III-V/Si DFB laser with a large relaxation oscillation frequency (ROF) of 14 GHz. The impact of short feedback on the ROF is studied and an increase of 2 GHz is observed. Under long feedback, the route to chaos of the device and its dependence on the short feedback dynamics are studied. The short feedback allows tuning the chaotic dynamics obtained under long feedback, and the increase of the ROF translates into an enhancement of the chaos bandwidth to above 16 GHz. This configuration can allow generation of wideband chaos using a single laser source in a photonic integrated circuit. (10.1109/JSTQE.2017.2732830)
  DOI : 10.1109/JSTQE.2017.2732830
• Analyse des traces d'usage de Gallica
  
  • Nouvellet Adrien
  • Beaudouin Valérie
  • d'Alché-Buc Florence
  • Prieur Christophe
  • Roueff François
  , 2017. Gallica est l'une des plus grandes bibliothèques numériques librement accessible sur le web. Dans le cadre du Bibli-Lab, partenariat de recherche entre la Bibliothèque nationale de France et Télécom ParisTech, et avec le soutien du TeraLab, a été conduite une analyse inédite des logs de connexion aux serveurs de Gallica, en leur appliquant des méthodes d’apprentissage automatique (machine learning). L’objectif n’était pas de connaître les usagers ni leurs profils mais, en partant de traces d’usages que sont les logs, d’identifier des parcours-types. Durant 15 mois (avril 2016-juillet 2017), un chercheur en contrat postdoctoral encadré par quatre enseignants-chercheurs de Télécom ParisTech , a mis au point un algorithme de classification (ou clusterisation) permettant de regrouper des sessions de Gallica présentant des similitudes dans l’enchaînement des actions. Les logs analysés couvraient des durées variables, allant d’une semaine à un mois, avec vérification systématique de la stabilité des modèles obtenus. Le choix méthodologique fort a été ici de faire dialoguer les modèles statistiques avec les résultats issus d’autres approches (observations ethnographiques, entretiens, etc. ). Ce dialogue a permis à la fois de : a) fixer les paramètres de départ (durée d’une session, définition des actions élémentaires sur Gallica) ; b) contrôler les modèles obtenus, extrêmement sensibles aux artefacts techniques ; c) proposer des premières clés d’interprétation.