Publications

2017

• On high dimensional regression: computational and statistical perspectives
  
  • Salmon Joseph
  , 2017. This dissertation essentially covers the work done by the author as a `Maître de Conférences'' at the Laboratoire de Traitement et Communication de l'Information (LTCI), at Télécom ParisTech, since December 2012. During this period, the author strengthened his contributions to high-dimensional statistics and in particular sparse regression methods. In particular, the main focus of the dissertation is on computational aspects and to speed-up algorithms for Lasso-type problems, on means to better take into account the unknown noise and on corrections against the bias non-smooth convex regression methods suffer from. This report is not meant to present comprehensive description of the results developed by the author, but rather a synthetic view of his main contributions. The interested reader may consult the referenced articles for additional details and more precise treatment of the topics presented here.
• Collaborative clustering based on Algorithmic Information Theory
  
  • Murena Pierre-Alexandre
  • Sublime Jérémie
  • Matei Basarab
  • Cornuéjols Antoine
  , 2017. Clustering is a compression task which consists in grouping similar objects into clusters. In real-life applications, the system may have access to several views of the same data and have each view process by a specific clustering algorithm: this framework is called collaborative clustering and can benefit from algorithms capable of exchanging information between the different views. In this paper, we consider this type of unsupervised ensemble learning as a compression problem and develop a theoretical framework based on algorithmic theory of information suitable for multi-view clustering and collaborative clustering applications. Using this approach, we propose a new algorithm based on solid theoretical basis, and test it on several real and artificial data sets.
• Copier-coller 3D : paramétrisation cohérente de maillages triangulaires
  
  • Vintescu Ana-Maria
  , 2017. Nous proposons un algorithme efficace pour le paramétrisation global de surfaces triangulées. Tout d’abord, les singularités coniques sont automatiquement détectées dans des endroits visuellement significatifs ; ce processus est efficace sur le plan calculatoire et vise à détecter de tels cônes aux sommets du maillage, où des valeurs élevées de distorsion de surface peuvent être prédites avant la paramétrisation réelle. Afin d’assurer la continuité au travers des coupes coniques résultant de la découpe du maillage, des fonctions de transition affines sont utilisées ; celles-ci sont intégrées dans un système linéaire qui vise à minimiser la distorsion angulaire. Dans cette thèse, nous présentons également un nouvel algorithme de paramétrisation croisé qui, étant donné deux maillages triangulaires d’entrée et des points de correspondance fournit par l’utilisateur, effectue des paramétrisations topologiquement et géométriquement cohérentes. La paramétrisation cohérente simultanée des maillages est réalisée en seulement quelques secondes, en résolvant au plus quatre systèmes linéaires au sens des moindres carrés. Nous validons les résultats des algorithmes proposés en fournissant des résultats expérimentaux étendus, en démontrant l’efficacité en temps, ainsi que la qualité, illustrée en examinant les mesures de distorsion acceptées. L’efficacité de calcul des algorithmes présentés permet leur utilisation dans des applications interactives, où l’utilisateur peut modifier ou ajouter des singularités coniques (ou des correspondances de référence pour le pipeline de paramétrisation croisé) tout en obtenant des résultats dans des temps de fonctionnement pratiques.
• Apprentissage distribué dans les jeux pour les réseaux sans fil
  
  • Ali Mohammed Shabbir
  , 2017. Dans cette thèse, nous abordons la question de la conception et de la gestion optimales des réseaux sans fil, qui sont devenus essentiels avec la demande croissante de données. Les approches centralisées deviennent impraticables, tandis que l'apprentissage distribué dans les jeux est une approche distribuée prometteuse. En particulier, les jeux potentiels sont attrayants car les joueurs du jeu peuvent optimiser de façon distribuée une fonction potentielle. Nous étendons cette notion aux jeux potentiels approchés et bruités et étudions la convergence des algorithmes d'apprentissage dans ces cadres. Nous appliquons ensuite les résultats théoriques obtenus aux réseaux sans fil. Dans une première partie, nous prouvons que, dans certaines conditions, l'algorithme d'apprentissage log-linéaire (LLA) et le LLA binaire (BLLA) convergent vers le minimum global de la fonction potentielle dans les jeux potentiels approchés. Nous prouvons alors la convergence de BLLA dans des jeux potentiels bruités pour des températures fixes et décroissantes. Nous fournissons un nombre suffisant d'échantillons d'estimation qui garantit la convergence pour un bruit borné ou non borné. Un facteur clé pour analyser la convergence des algorithmes proposés est la résistance des arbres dans une Chaîne de Markov Perturbée (PMC). Nous développons de nouvelles règles pour simplifier le calcul de la résistance des arbres. Dans une deuxième partie, nous appliquons nos résultats à des problèmes pratiques dans les réseaux sans fil. Nous considérons d'abord l'équilibrage de charge dans des réseaux cellulaires hétérogènes. Dans ce contexte, nous proposons également un nouveau schéma de recuit pour LLA adapté aux horizons finis. Ensuite, nous appliquons BLLA au problème d'assignation de canal (CAP) dans les réseaux de communication de terminal à terminal (D2D). En raison du bruit d'estimation des débits, le cadre de jeu à potentiel bruité se pose naturellement. Nous abordons enfin un problème de maximisation de fonction sous modulaire avec contraintes. Nous le concevons comme un jeu fini avec des joueurs ayant des informations limitées ou inexistantes sur leur voisinage. Nous caractérisons la performance de l'algorithme glouton et nous fournissons le meilleur graphe d'information pour un nombre donné de joueurs et d'arêtes d'information.
• Protocol for preserving privacy in distributed system (PPDS)
  
  • Aloui Ashref
  • Msahli Mounira
  • Abdessalem Talel
  • Bressan Stephane
  • Mesnager Sihem
  , 2017, pp.1885-1890. (10.1109/IWCMC.2017.7986571)
  DOI : 10.1109/IWCMC.2017.7986571
• Scalable energy efficient routing in multi-layer femtocell networks
  
  • Zhang Jun
  • Labiod Houda
  • Hammami Seif Eddine
  • Afifi Hossam
  , 2017, pp.1540 - 1545. In this paper, we present a data-driven solution to address the energy efficient routing problem in a multi-layer femtocell network. The basic idea is to divide problems into categories by the clustering of traffic demand vectors, and solve a problem by a most suitable algorithm, in terms of throughput, energy consumption and computation time, corresponding to its category. The simulation results show that, our proposed approach can achieve a similar performance in throughput and energy saving as the previous JRN algorithm, by reducing 80% computation time (10.1109/IWCMC.2017.7986513)
  DOI : 10.1109/IWCMC.2017.7986513
• Synthèse de vues et reconstruction de vues à partir de vidéos compressées multi-vues et multi-sources
  
  • Purica Andrei
  , 2017. De nos jours, les vidéos sont la forme de multimédia la plus demandée. Ce grand intérêt a alimenté une évolution continue des technologies d’affichage, de transmission et de compression vidéo. Également, il y a aussi beaucoup d’intérêt à trouver le meilleur moyen d’offrir une expérience multimédia dite immersive. Plusieurs solutions ont été étudiées au cours des dernières années et le format vidéo multi-vues plus profondeur a été trouvé pour fournir une solution prometteuse en combinaison avec des algorithmes de synthèse visuelle. Dans cette thèse, nous explorons plusieurs sujets liés à la synthèse et à la reconstruction des vues. Tout d’abord, nous explorons l’utilisation des corrélations temporelles en combinaison avec les techniques traditionnelles de rendu basé sur la profondeur d’image et proposons plusieurs approches pour aborder les problèmes communs des algorithmes de ce type qui sont démontrés pour améliorer la qualité de la synthèse. Comme les algorithmes de synthèse de vues produisent des distorsions localisées élevées, nous évaluons également l’efficacité des mesures d’évaluation de qualité courantes et proposons une évaluation ciblée sur la région d’intérêt. Enfin, nous étudions le problème de la reconstruction vidéo multisource et proposons un modèle de reconstruction qui utilise des algorithmes proximaux primal-dual d’optimisation convexes pour améliorer la qualité et la résolution des vidéos provenant de sources multiples avec des résolutions et des niveaux de compression éventuellement différents.
• Rate-distortion regions of instances of cascade source coding with side information
  
  • Wang Chien-Yi
  • Zaidi Abdellatif
  , 2017, pp.1918-1922.
• Bounded Network Coding Redundancy for Multi-Layer Video Streaming
  
  • Baccouch Hana
  • Paul-Louis Ageneau
  • Tizon Nicolas
  • Boukhatem Nadia
  • Nguyen Thi-Mai-Trang
  , 2017. In this paper, we address the tradeoff between network coding overhead and multi-layer video tolerated loss rate on multi-hop wireless networks. We aim, through a simulation study, to evaluate a bounded network coding redundancy, for scalable video, based on the measured packet loss and the targeted maximum application loss rate. We consider, first, applying the bounded redundancy on a generation based scheme, and evaluate the tradeoff between tolerated video loss and the introduced overhead. Second, since scalable video produces a high video quality with various levels of importance, we propose to apply the bounded redundancy to an Expanding Window based scheme. We carried out a performance evaluation of both network coding schemes with two video test sequences using ns-3 simulator. We show the benefits of our scheme particularly in terms of overhead and PSNR.
• A Lightweight Mutual Authentication Protocol for the IoT
  
  • Hammi Mohamed Tahar
  • Livolant Erwan
  • Bellot Patrick
  • Serhrouchni Ahmed
  • Minet Pascale
  , 2017. The Internet of Things enables the interconnection of smart physical and virtual objects, managed by highly developed technologies. WSN, is an essential part of this paradigm. The WSN uses smart, autonomous and usually limited capacity devices in order to sense and monitor industrial environments. However, if no authentication mechanism is deployed, this system can be accessible, used and controlled by non-authorized users. In this paper, we propose a robust WSN mutual authentication protocol. A real implementation of the protocol was realized on OCARI, one of the most interesting Wireless Sensor Network technologies. All nodes wanting to access the network should be authenticated at the MAC sub-layer of OCARI. This protocol is especially designed to be implemented on devices with low storage and computing capacities. (10.1007/978-981-10-5281-1)
  DOI : 10.1007/978-981-10-5281-1
• Asymptotic Capacity Results for MIMO Wireless Optical Communication
  
  • Moser Stefan M
  • Mylonakis Michail M
  • Wang Ligong
  • Wigger Michèle
  , 2017. This paper provides several asymptotic capacity results for the multiple-input multiple-output free-space optical intensity channel in the regime of high signal-to-noise ratio (SNR). For the case where the channel matrix has full column rank, the asymptotic capacity is derived assuming a peak-power constraint on each transmit antenna, or an average-power constraint on the total power across all transmit antennas, or both. For multiple-input and single-output channels, the asymptotic high-SNR capacity is derived when either only the total average power is constrained, or only the per-antenna peak power is constrained, or both but with the average-power constraint being sufficiently loose.
• An FIR Memory Polynomial Predistorter for Wideband RF Power Amplifiers
  
  • Manyam Venkata Narasimha
  • Pham Dang-Kièn Germain
  • Jabbour Chadi
  • Desgreys Patricia
  , 2017, pp.249-252. (10.1109/NEWCAS.2017.8010152)
  DOI : 10.1109/NEWCAS.2017.8010152
• Correlation engineering for a flexible Raman free fibered photon source
  
  • Cordier Martin
  • Orieux Adeline
  • Gabet Renaud
  • Harlé Thibault
  • Dubreuil Nicolas
  • Diamanti Eleni
  • Delaye Philippe
  • Zaquine Isabelle
  , 2017. Fibered sources of photon pairs can be easily integrated into future quantum communication networks. However, in silica-core fibres, the broadband spectrum of spontaneous Raman scattering strongly contributes to uncorrelated noise photons degrading the quality of the source. To overcome the problem of the silica Raman scattering, a new architecture was recently proposed, with hollow-core photonic crystal fibers (LF-HCPF) whose core and cladding are filled with a non-linear liquid. As opposed to the Raman scattering spectrum of silica, which is broadband and continuous, the Raman scattering spectrum of liquids is composed of narrow lines. Thus, generating photon pairs between these narrow lines permits to avoid the spectral overlap between the SFWM pairs and the Raman photons. A near infra-red LF-HCPCF photon pair source has been demonstrated, featuring a three order of magnitude suppression of the Raman noise [1]. We have also demonstrated the first non-linear LF-PCF with a transmission band and a zero-dispersion-wavelength (ZDW) in the telecom wavelength range combined with a non-linearity of the same order of magnitude as the one of silica, as a first step towards a Raman-free photon pair source in the telecom band [2]. Injecting light on both sides of this fiber in a Sagnac loop configuration can then be used to obtain entanglement. (10.1109/CLEOE-EQEC.2017.8087344)
  DOI : 10.1109/CLEOE-EQEC.2017.8087344
• Reducing Phonon-Induced Decoherence in Solid-State Single-Photon Sources with Cavity Quantum Electrodynamics
  
  • Grange T.
  • Somaschi N.
  • Anton C.
  • de Santis L.
  • Coppola G.
  • Giesz V.
  • Lemaitre A.
  • Sagnes I.
  • Auffèves A.
  • Senellart P.
  Physical Review Letters, American Physical Society, 2017, 118 (25), pp.253602. (10.1103/PhysRevLett.118.253602)
  DOI : 10.1103/PhysRevLett.118.253602
• On Achievability for Downlink Cloud Radio Access Networks with Base Station Cooperation.
  
  • Wang Chien-Yi
  • Wigger Michèle
  • Zaidi Abdellatif
  , 2017, pp.1-6.
• Comparison of distances for supervised segmentation of white matter tractography
  
  • Olivetti Emanuele
  • Bertò Giulia
  • Gori Pietro
  • Sharmin Nusrat
  • Avesani Paolo
  , 2017. Tractograms are mathematical representations of the main paths of axons within the white matter of the brain, from diffusion MRI data. Such representations are in the form of polylines, called streamlines, and one streamline approximates the common path of tens of thousands of axons. The analysis of tractograms is a task of interest in multiple fields, like neurosurgery and neurology. A basic building block of many pipelines of analysis is the definition of a distance function between streamlines. Multiple distance functions have been proposed in the literature, and different authors use different distances, usually without a specific reason other than invoking the " common practice ". To this end, in this work we want to test such common practices, in order to obtain factual reasons for choosing one distance over another. For these reason, in this work we compare many streamline distance functions available in the literature. We focus on the common task of automatic bundle segmentation and we adopt the recent approach of supervised segmentation from expert-based examples. Using the HCP dataset, we compare several distances obtaining guidelines on the choice of which distance function one should use for supervised bundle segmentation. (10.1109/PRNI.2017.7981502)
  DOI : 10.1109/PRNI.2017.7981502
• A Survey on Ensemble Learning for Data Stream Classification
  
  • Gomes Heitor Murilo
  • Barddal Jean Paul
  • Enembreck Fabrició
  • Bifet Albert
  ACM Computing Surveys, Association for Computing Machinery, 2017, 50 (2), pp.1 - 36. (10.1145/3054925)
  DOI : 10.1145/3054925
• Formalism to assess the entropy and reliability of the loop-PUF
  
  • Danger Jean-Luc
  • Rioul Olivier
  • Guilley Sylvain
  • Schaub Alexander
  , 2017. Many Methods exist for Physically Unclonable Function in order to enhance their reliability, but they are at the cost of extra hard- ware, either coming from the PUF core itself, or from the correction method, as fuzzy extraction [5]. Another issue is to better know the PUF entropy, especially when the PUF is used as a cryptographic key generator. The SRAM and Ring Oscillator PUF provide relatively high entropy, as shown experimentally in [4]. The min-entropy is gen- erally better assessed, as studied in Delvaux et al. [2] which presents the min-entropy of strong PUFs. For strong PUFs, it appears that the number of required challenges to get the min-entropy is signif- ficantly higher, thus involving compression and extra complexity to generate key bits. We show in this study that it is possible to ob- tain a strong PUF, relying on a Loop PUF, which presents a high ratio between reliability and complexity. Moreover the entropy can be known when choosing the minimal and optimal number of chal- lenges, namely Hadamard Codes. As the presence of noise inevitably involves an entropy decrease, this paper presents a method allowing the PUF to compensate this loss to meet the requires entropy. The single Ring-Oscillator of the Loop PUF, the hard-coded challenges and the lightweight error correction provide a low complexity PUF, around 1000 μm2 , with an access time of 20 ms to get a 64 key bits with a Bit Error Rate (BER) which can be less than 10−9 by filter- ing unreliable bits. As the challenges and responses remain local, the modeling attacks are signiffificantly reduced and are not considered in this work.
• Adaptive random forests for evolving data stream classification
  
  • Bifet Albert
  • Gomes Heitor
  • Read Jesse
  • Barddal Jean Paul
  • Enembreck Fabrició
  • Pfharinger Bernhard
  • Holmes Geoff
  • Abdessalem Talel
  Machine Learning, Springer Verlag, 2017, 106 (9-10), pp.1469 - 1495. Random forests is currently one of the most used machine learning algorithms in the non-streaming (batch) setting. This preference is attributable to its high learning performance and low demands with respect to input preparation and hyper-parameter tuning. However, in the challenging context of evolving data streams, there is no random forests algorithm that can be considered state-of-the-art in comparison to bagging and boosting based algorithms. In this work, we present the adaptive random forest (ARF) algorithm for classification of evolving data streams. In contrast to previous attempts of replicating random forests for data stream learning, ARF includes an effective resampling method and adaptive operators that can cope with different types of concept drifts without complex optimizations for different data sets. We present experiments with a parallel implementation of ARF which has no degradation in terms of classification performance in comparison to a serial implementation, since trees and adaptive operators are independent from one another. Finally, we compare ARF with state-of-the-art algorithms in a traditional test-then-train evaluation and a novel delayed labelling evaluation, and show that ARF is accurate and uses a feasible amount of resources. (10.1007/s10994-017-5642-8)
  DOI : 10.1007/s10994-017-5642-8
• Directed Acyclic Graph Scheduling for Mixed-Criticality Systems
  
  • Medina Roberto
  • Borde Etienne
  • Pautet Laurent
  , 2017, pp.217-232.
• Stochastic Side-Channel Leakage Analysis <i>via</i> Orthonormal Decomposition
  
  • Guilley Sylvain
  • Heuser Annelie
  • Tang Ming
  • Rioul Olivier
  , 2017, 10543. Side-channel attacks of maximal efficiency require an accurate knowledge of the leakage function. Template attacks have been introduced by Chari et al. at CHES 2002 to estimate the leakage function using available training data. Schindler et al. noticed at CHES 2005 that the complexity of profiling could be alleviated if the evaluator has some prior knowledge on the leakage function. The initial idea of Schindler is that an engineer can model the leakage from the structure of the circuit. However, for some thin CMOS technologies or some advanced countermeasures, the engineer intuition might not be sufficient. Therefore, inferring the leakage function based on profiling is still important. In the state-of-the-art, though, the profiling stage is conducted based on a linear regression in a non-orthonormal basis. This does not allow for an easy interpretation because the components are not independent. In this paper, we present a method to characterize the leakage based on a Walsh-Hadamard orthonormal basis with staggered degrees, which allows for direct interpretations in terms of bits interactions. A straightforward application is the characterization of a class of devices in order to understand their leakage structure. Such information is precious for designers and also for evaluators, who can devise attack bases relevantly.
• Characterizing the maximum parameter of the total-variation denoising through the pseudo-inverse of the divergence
  
  • Deledalle Charles-Alban
  • Papadakis Nicolas
  • Salmon Joseph
  • Vaiter Samuel
  , 2017. We focus on the maximum regularization parameter for anisotropic total-variation denoising. It corresponds to the minimum value of the regularization parameter above which the solution remains constant. While this value is well know for the Lasso, such a critical value has not been investigated in details for the total-variation. Though, it is of importance when tuning the regularization parameter as it allows fixing an upper-bound on the grid for which the optimal parameter is sought. We establish a closed form expression for the one-dimensional case, as well as an upper-bound for the two-dimensional case, that appears reasonably tight in practice. This problem is directly linked to the computation of the pseudo-inverse of the divergence, which can be quickly obtained by performing convolutions in the Fourier domain.
• A Correlation-based Dissimilarity Measure for Noisy Patches
  
  • Riot Paul
  • Almansa Andrés
  • Gousseau Yann
  • Tupin Florence
  , 2017, 10302, pp.184-195. In this work, we address the problem of defining a robust patch dissimilarity measure for an image corrupted by an additive white Gaussian noise. The whiteness of the noise, despite being a common assumption that is realistic for RAW images, is hardly used to its full potential by classical denoising methods. In particular, the $\ell^2$-norm is very widely used to evaluate distances and similarities between images or patches. However, we claim that a better dissimilarity measure can be defined to convey more structural information. We propose to compute the dissimilarity between patches by using the autocorrelation of their difference. In order to illustrate the usefulness of this measure, we perform three experiments. First, this new criterion is used in a similar patch detection task. Then, we use it on the Non Local Means (NLM) denoising method and show that it improves performances by a large margin. Finally, it is applied to the task of no-reference evaluation of de-noising results, where it shows interesting visual properties. In all those applications, the autocorrelation improves over the $\ell^2$-norm. (10.1007/978-3-319-58771-4_15)
  DOI : 10.1007/978-3-319-58771-4_15
• PROCEDE DE GESTION D'UN ENSEMBLE D'APPAREILS CONSOMMATEURS D'ENERGIE ELECTRIQUE, ET MODULE GESTIONNAIRE D'ENERGIE ELECTRIQUE
  
  • Al Zahr Sawsan
  • Rousselle Mélaine
  • Forestier Philippe
  , 2017.
• Streaming And Presentation Architectures For Extended Video Streams
  
  • Potetsianakis Emmanouil
  , 2017. Mobile devices able to capture multimedia streams (audio and video) have now the capability to produce numerous associated data (GPS, Gyroscope, etc.). The real-time exploitation of these extra-data, in different application scenarios, raise the issue of integrating these data in existing broadcasting architectures. These architectures are designed to work with the audiovisual data and should be extended to match the extradata requirements (processing time, bandwidth, visualization techniques etc.), and be able to adapt to their environment (network, device) accordingly. The major part of the applications utilizing these data are developed in an &quot;ad-hoc&quot; way, without taking all of their characteristics in account, and require complex maintenance and update efforts in case new data types or new network/device support is introduced. Our study is focused on optimizing the present distribution and presentation architectures for Audiovisual (AV) content, in order to find an efficient way to represent the associated data. (10.1145/3084289.3084298)
  DOI : 10.1145/3084289.3084298