Publications

2019

• Recurring concept meta-learning for evolving data streams
  
  • Anderson Robert
  • Koh Yun Sing
  • Dobbie Gillian
  • Bifet Albert
  Expert Systems with Applications, Elsevier, 2019, 138, pp.112832. When concept drift is detected during classification in a data stream, a common remedy is to retrain a framework’s classifier. However, this loses useful information if the classifier has learnt the current concept well, and this concept will recur again in the future. Some frameworks retain and reuse classifiers, but it can be time-consuming to select an appropriate classifier to reuse. These frameworks rarely match the accuracy of state-of-the-art ensemble approaches. For many data stream tasks, speed is important: fast, accurate frameworks are needed for time-dependent applications. We propose the Enhanced Concept Profiling Framework (ECPF), which aims to recognise recurring concepts and reuse a classifier trained previously, enabling accurate classification immediately following a drift. The novelty of ECPF is in how it uses similarity of classifications on new data, between a new classifier and existing classifiers, to quickly identify the best classifier to reuse. It always trains both a new classifier and a reused classifier, and retains the more accurate classifier when concept drift occurs. Finally, it creates a copy of reused classifiers, so a classifier well-suited for a recurring concept will not be impacted by being trained on a different concept. In our experiments, ECPF classifies significantly more accurately than a state-of-the-art classifier reuse framework (Diversity Pool) and a state-of-the-art ensemble technique (Adaptive Random Forest) on synthetic datasets with recurring concepts. It classifies real-world datasets five times faster than Diversity Pool, and six times faster than Adaptive Random Forest and is not significantly less accurate than either. (10.1016/j.eswa.2019.112832)
  DOI : 10.1016/j.eswa.2019.112832
• Stochastic Dosimetry for Radio-Frequency Exposure Assessment in Realistic Scenarios
  
  • Chiaramello E.
  • Fiocchi S.
  • Parazzini M.
  • Ravazzani P.
  • Wiart Joe
  , 2019, pp.89-102. (10.1007/978-3-030-04870-9_6)
  DOI : 10.1007/978-3-030-04870-9_6
• Sparse channels estimation applied in software defined radio
  
  • Crespo Marques Elaine
  , 2019. Communication channels are used to transmit information signals. However, these channels can cause several distortions on the signal to be transmitted, such as attenuation, multipath loss and Doppler shift, among others. For a better message recovery, the receiver can estimate the channel and bring more reliability to the communications systems. Several communications systems, for example high-definition television, mmWave system, wideband HF and ultra-wideband have sparse channels. This characteristic can be used to improve the performance of the estimator and reduce the size of the training sequence so decreasing the consumption power and bandwidth. This thesis handles the channel estimation problem by investigating methods that exploit the sparsity of the channel. The study of Compressive Sensing and its sparse recovery algorithms led to the proposition of a new algorithm called Matching Pursuit based on Least Square (MPLS). The use of neural networks (NN) to sparse signals estimation was also explored. The work focused on NN inspired by sparse recovery algorithms such as Learned Iterative Shrinkage-Thresholding Algorithm (LISTA). This resulted in two approaches that improve LISTA performance as well as to a new neural network suitable to estimate sparse signals.
• An Analysis of the Effect of Data Augmentation Methods: Experiments for a Musical Genre Classification Task
  
  • Mignot Rémi
  • Peeters Geoffroy
  Transactions of the International Society for Music Information Retrieval (TISMIR), Ubiquity Press, 2019, 2 (1), pp.97-110. (10.5334/tismir.26)
  DOI : 10.5334/tismir.26
• Recursive multivariate derivatives of $e^{f(x_1,\dots, x_n)}$ of arbitrary order
  
  • Miatto Filippo M
  • Miatto Filippo M.
  , 2019. High-order derivatives of nested functions of a single variable can be computed with the celebrated Fa\`a di Bruno's formula. Although generalizations of such formula to multiple variables exist, their combinatorial nature generates an explosion of factors, and when the order of the derivatives is high, it becomes very challenging to compute them. A solution is to reuse what has already been computed, which is a built-in feature of recursive implementations. Thanks to this, recursive formulas can play an important role in Machine Learning applications, in particular for Automatic Differentiation. In this manuscript we provide a recursive formula to compute multivariate derivatives of arbitrary order of $e^{f(x_1,\dots,x_n)}$ with respect to the variables $x_i$. We note that this method could also be beneficial in cases where the high-order derivatives of a function $f(x_1,\dots,x_n)$ are hard to compute, but where the derivatives of $\log(f(x_1,\dots,x_n))$ are simpler.
• Graphical Calculus for products and convolutions
  
  • Miatto Filippo M
  , 2019. Graphical calculus is an intuitive visual notation for manipulating tensors and index contractions. Using graphical calculus leads to simple and memorable derivations, and with a bit of practice one can learn to prove complex identities even without the need for pen and paper. This manuscript is meant as a demonstration of the power and flexibility of graphical notation and we advocate exploring the use of graphical calculus in undergraduate courses. In the first part we define the following matrix products in graphical language: dot, tensor, Kronecker, Hadamard, Kathri-Rao and Tracy-Singh. We then use our definitions to prove several known identities in an entirely graphical way. Despite ordinary proofs consist in several lines of quite involved mathematical expressions, graphical calculus is so expressive that after writing an identity in graphical form we can realise by visual inspection that it is in fact true. As an example of the intuitiveness of graphical proofs, we derive two new identities. In the second part we develop a graphical description of convolutions, which is a central ingredient of convolutional neural networks and signal processing. Our single definition includes as special cases the circular discrete convolution and the cross-correlation. We illustrate how convolution can be seen as another type of product and we derive a generalised convolution theorem. We conclude with a quick guide on implementing tensor contractions in python.
• TDD based 5G networks performance analysis and radio resource dimensioning with different network geometry models
  
  • Rachad Jalal
  , 2019. A new generation of cellular networks, known as 5G new radio (NR), has been standardized through improvements in the current 4G/4G+ concepts and features in order to bring a new level of flexibility, scalability and efficiency. Time division duplex (TDD) is expected to be one of the key features of 5G NR since it offers more advantages than frequency division duplex (FDD) mode in terms of capacity, flexibility and implementation adequacy with other features, such as full dimension multiple input multiple output antenna (FD-MIMO) technology. A variant operational mode of TDD, known as D-TDD, is in the scope. It is designed to deal with uplink (UL) and downlink (DL) traffic asymmetry since it is based on instantaneous traffic estimation and offers more flexibility in resource assignment. However, the use of D-TDD requires new interference mitigation schemes capable to handle two additional types of interference called cross link interference (CLI) and stands for DL to UL and UL to DL interference. The first part of this thesis is devoted to tackle the problem of interference modeling in D-TDD based macro-cell and small-cell deployments. We provide a complete analytical approach to derive relevant metrics, such as interference-to-signal-ratio (ISR) and the coverage probability formulas, considering adequate geometry models for each type of deployment. Then, we propose an interference mitigation scheme based on 3D beamforming for macro-cells and a cell-clustering scheme for small-cells. Additionally, we investigate another problematic which is Orthogonal Frequency Division Multiplex (OFDM) radio resource dimensioning. Since 5G NR takes in consideration a wide array of emerging use cases and also the possibility of having future requirements, the third Generation Partnership Project (3GPP) comes up with a variant of OFDM known as scalable OFDM and having different sub carriers' spacing. This feature appears to be an ideal choice for 5G NR since it offers a high spectral efficiency, robustness to selective fading channels, convenience with diverse spectrum bands and compatibility with other features. Therefore, dimensioning OFDM is a major task to accomplish in the context of NR. To this purpose, we provide an analytical model to dimension OFDM based systems with a proportional fair resources' allocation policy. We model indoor users by a spatial PPP and outdoor users according to a Cox process driven by PLP. Different analytical and numerical results are provided to justify the accuracy of this model.
• Towards a scalable and programmable incremental deployment of ICN in the real world
  
  • Sardara Mauro
  , 2019. Information-Centric Networking (ICN) embraces a family of network architectures rethinking Internet communication principles around named-data. After several years of research and the emergence of a few popular proposals, the idea to replace TCP/IP with data-centric networking remains a subject of debate. ICN advantages have been advocated in the context of 5G networks for the support of highly mobile, multi-access/source and latency minimal patterns of communications. However, large-scale testing and insertion in operational networks are yet to happen, likely due to the lack of a clear incremental deployment strategy. The aim of this thesis is to propose and evaluate effective solutions for deploying ICN. Firstly, we propose Hybrid-ICN (hICN), an ICN integration inside IP (rather that over/ under/ in place of) that has the ambition to trade-off no ICN architectural principles. By reusing existing packet formats, hICN brings innovation inside the IP stack, requiring minimal software upgrades and guaranteeing transparent interconnection with existing IP networks. Secondly, the thesis focuses on the problem of deploying ICN at the network endpoints, namely at the end host, by designing a transport framework and a socket API that can be used in several ICN architectures such as NDN, CCN and hICN. The framework fosters cutting-edge technologies aiming at providing performance and efficiency to applications. An extensive benchmarking at the end of the chapter will present the performance of the transport framework. Subsequently, the benefits that hICN network and transport services can bring to applications will be assessed, by considering two main use cases: HTTP and WebRTC. The former represents the de-facto protocol of the Web, while the latter is a new emerging technology increasingly adopted for real time services. At last, the thesis proposes a solution for programmatically deploying, configuring and managing ICN networks and applications: Virtualized ICN (vICN), a programmable unified framework for network configuration and management that uses recent progresses in resource isolation and virtualization techniques. It offers a single, flexible and scalable platform to serve different purposes, in particular the real deployments of ICN in existing IP networks.
• Reconfigurable Lattice Agreement and Applications
  
  • Kuznetsov Petr
  • Rieutord Thibault
  • Sara Tucci-Piergiovanni
  , 2019.
• Interference management in sectored cellular networks with local multi-cell processing
  
  • Gelincik Samet
  , 2019. The data rate requirement in wireless communication due to employment of smartphones, laptops, tablets and sensors is increasing drastically. This directly poses extra-ordinary demands on precious spectral resources. To satisfy with the expected saturation on the currently used bands, modern communication systems are allowing very aggressive spatial frequency reuse and moving towards heterogenous networks of base stations (BS) covering smaller areas (small cells). Evidently, such system suffer from the detrimental inter-cell interference conditions, particularly at cell edges. Therefore, it is clearly convincing that interference management is a bottleneck for current and future wireless networks. Multi-cell processing (MCP) schemes has mostly been used to provide BSs with quantized versions of the transmit/receive signals of other BSs via backhaul/fronthaul links (allowing for clustered decoding). It is then possible for user data to be jointly processed by several BSs at both uplink and downlink, hence imitating the benefits of virtual MIMO. However, the implementation of MCP for all the BSs of the network is quite challenging in practice due to large computational complexity and excessive delays even for moderately large networks. However, dividing the network into several clusters and letting them to cooperate within each cluster rather than the entire network also brings some benefit of MCP by requiring only local received signals and local CSI. We name this framework as local MCP, which also improves the robustness of the network to connection failures and scalability. In this thesis, we have investigated the benefits of local MCP in interference management for sectored hexagonal network model under three different scenarios. In the first one, we assumed that the BS can cooperate through limited capacity links for a given number of cooperation rounds. We proposed a new practical clustering scheme that adapts the way BSs cooperate to cells sectorization. Upper and lower bounds on the peruser degrees-of-freedom (DoF) as a function of number cooperation round and backhaul capacity have been derived, and finite SNR analysis has been done. In the second scenario, we assumed a multi-cloud cellular system, where each central processor (CP) has a limited processing power. A clustering scheme has been proposed that adapts the association between BSs and CPs to the sectorization. Lower bound on the per-user DoF as a function of fronthaul capacity, CP capacity and the ratio of number of CP to number of BS has been derived. In the last scenario, we assumed again a multi-cloud based cellular system, and applied compute-and-forward (CoF) and Quantized CoF schemes to the proposed clustering. For Quantized CoF, we proposed a method for reducing the number of nested lattice codes to lower the implementation complexity while keeping reasonable performance degradation.
• Augmentation de la capacité des interfaces PONs TDMA au-delà de 10Gbit/s dans les réseaux d’accès en fibres optiques
  
  • Barthomeuf Sylvain
  , 2019. Le réseau d’accès optique est en plein essor, notamment avec le déploiement massif de Fiber to-the-home (FTTH) amenant la fibre jusqu’à l’usager et l’arrivée de la 5ème génération (5G) pour les communications mobiles. Sa démocratisation et l’évolution des usages d’Internet encourage la montée en débit au-delà des 10Gbit/s prévus avec le XG(S)-PON. Il existe de nombreuses solutions pour y parvenir. Dans cette thèse, la montée en débit du réseau d’accès optique passif (PON) sera étudiée sous trois axes :• L’utilisation du format de modulation PAM4 pour réutiliser les composants optiques et électrique sà 10GHz pour atteindre 25Gbit/s. Ce format de modulation à quatre niveaux d’amplitude, en plus d’avoir une meilleure efficacité spectrale que le NRZ couramment utilisé, permet l’introduction d’une structure de PON fonctionnant à deux débits. Au cours du chapitre traitant du PAM4, l’amplification optique sera étudiée pour améliorer les performances de budget optique. La dernière partie traite de l’extrapolation des résultats expérimentaux pour simuler une transmission PAM8.• Les propositions des industriels concernant la montée en débit, à travers trois prototypes. Le premier prototype concerne une application de la norme NG-PON2 et de son option PtP WDM. Le deuxième prototype démontre la faisabilité d’une transmission duobinaire à 25Gbit/s en bande latérale résiduelle. Le troisième et dernier prototype présentera une ébauche de solution pour réaliser le lien fronthaul pour la 5G mobile avec une technologiePtP WDM à 25Gbit/s par canal.• L’implémentation de l’égalisation dans le cadre du PON. Le traitement du signal est très utilisé dans les transmissions optiques du réseau de transport. Avec la montée en débit, son utilisation dans le réseau d’accès semble inévitable. Dans cette partie, les spécificités du PON et des transmissions point-à-multipoints sont étudiées pour proposer une implémentation d’égalisation en accord avec les contraintes de coûts du réseau d’accès et les contraintes liées à la topologie du réseau d’accès.
• Simulations et essais cliniques virtuels pour l'évaluation de la valeur clinique de l'angio-tomosynthèse par rapport à l'angio-mammographie
  
  • Sanchez de La Rosa Ruben
  , 2019. Contrast Enhanced Spectral Mammography (CESM) and Contrast Enhanced Digital Breast Tomosynthesis (CEDBT) sont des techniques d'imagerie par rayons X à double énergie impliquant l'injection d'un agent de contraste vasculaire. Les deux techniques fournissent des informations sur l'hypervascularisation des lésions par l'absorption de contraste. CESM est une application d'imagerie récemment introduite offrant un meilleur diagnostic du cancer du sein que la mammographie de diagnostique. CEDBT est une technique prometteuse fournissant des informations en trois dimensions, ce qui peut atténuer les limites du CESM. Cependant, sa valeur clinique par rapport à l'examen actuel CESM n'est toujours pas bien comprise. Notre travail vise à évaluer les performance de chaque technique afin de détecter et de caractériser avec précision les différentes absorptions de contraste apparaissant lors des examens cliniques. Cependant, une telle évaluation nécessiterait une vaste base de données de données cliniques avec une représentation complète des divers captages de contraste insérés dans différents contextes hétérogènes, ainsi que la vérité sur le terrain de chaque cas. Au lieu de cela, la nouveauté de notre proposition réside dans l'évaluation des deux techniques à l'aide de données simulées réalistes. Premièrement, nous avons amélioré le réalisme d’un outil de simulation de rayons X analytique. Ensuite, nous avons proposé un nouveau modèle d’absorption de contraste en masse, liée aux lésions mammaires et d’absorption naturelle de contraste de l’anatomie du sein. Enfin, nous avons réalisé un essai clinique virtuel évaluant les performances de détectabilité et de caractérisation du CESM et de la CEDBT à l'aide des outils de simulation développés dans notre travail. Les résultats obtenus montrent que la CEDBT offre une certaine valeur ajoutée clinique par rapport au CESM. En tant que travail futur, les résultats cliniques attendus grâce à la plate-forme de simulation proposée doivent être comparés aux vraies évaluations cliniques lorsqu'un système réel sera disponible.
• Multi-évaluation du Pointage : Application à la Dyspraxie
  
  • Gori Julien
  • Beaudouin-Lafon Michel
  • Guiard Yves
  , 2019 (4), pp.1--12. La loi de Fitts est utilisée invariablement en IHM pour évaluer la performance du pointage. Elle est toutefois confrontée à certains défauts, qui peuvent être exacerbés lors d'évaluations avec des populations atypiques. Dans cet article, nous présentons trois outils d'évaluation du pointage proposées récemment (le modèle WHo, la régression EMG et la méthode des PVPs) qui fournissent des réponses à ces défauts. Nous les illustrons sur des données réelles (4 sujets typiques, 2 jeunes dyspraxiques) et discutons des avantages et inconvénients de chaque méthode de cette nouvelle boite à outils. (10.1145/3366550.3372250)
  DOI : 10.1145/3366550.3372250
• Une architecture pour améliorer la réutilisatibilté des interfaces graphiques
  
  • Lecolinet Eric
  , 2019, pp.13:1-8. La programmation des interfaces graphiques reste un exercice laborieux et consommateur en temps l'essentiel de l'interaction devant généralement être codé au moyen d'un langage de programmation. Ce code, souvent verbeux et peu lisible, impacte la réutilisabilité des interfaces et la conception itérative, la moindre modification de l'interface nécessitant qu'il soit substantiellement modifié. Nous proposons un modèle architectural et une boîte à outils expérimentale qui facilitent les modifications et en les rendant peu coûteuses. (10.1145/3366551.3370352)
  DOI : 10.1145/3366551.3370352
• Study of solutions for automatic reconfiguration of future transport networks based on flexible optical layer
  
  • Alahdab Mhd Luay
  , 2019. Expanding the transport optical network using equipment supplied from different vendors, or what is called equipment interoperability, reduces the network cost, increases the quality and accelerates the delivery of network services. For that we study in this dissertation the requirements to apply the interoperability over the optical transport network. We focus on one particular case, called the alien wavelength use case, in which a WDM lightpath can be setup using a pair of third party transponders over any WDM optical line. The alien wavelength use case becomes manually doable, yet automating the alien lightpath setup process remains challenging due to the following reasons: I) The lack of a common communication between the third party transponders and the optical line prevents configuring the equipment properly. II) Most of the transponders’ optical parameters are propitiatory and/ or not standardized, and exchanging these parameters is required to estimate the physical performance of the path. III) Moreover, vendors’ estimation tools calculate the performance of the lightpath differently. The 1st part of this dissertation presents and analyzes most of the proposed approaches, including ours namely ”RSVP-TEbased approach”, that automate the alien lightpath setup process by providing a common or standardized communication and exchange the required transponders’ optical parameters. A comparison between the proposed approaches is made to find out the attributes of the ideal solution. The 2nd part of this dissertation focuses on upgrading the RSVP-TE-based approach to support flex-grid technology and configurable transponders that requires transponders’ configuration matching. For that we propose a standardized method, based on the RSVP-TE protocol, to fulfill that matching. The previous proposal has been demonstrated over the testbed of Orange labs and submitted as an IETF RSVP-TE standard draft as well. We also propose another standardized method, based on the OSPF-TE protocol, to exchange the raw optical parameters of the optical line equipment. Since exchanging raw optical parameters permits the proper use of vendor-agnostic physical performance estimation tools. The previous proposal has been demonstrated over the testbed and submitted as an IETF OSPF-TE standard draft. The 3rd part of this dissertation presents and demonstrates the module we developed to compute the path over a flexible network.
• Security and implementation of advanced quantum cryptography : quantum money and quantum weak coin flipping
  
  • Bozzio Mathieu
  , 2019. Harnessing the laws of quantum theory can drastically boost the security of modern communication networks, from public key encryption to electronic voting and online banking. In this thesis, we bridge the gap between theory and experiment regarding two quantum-cryptographic tasks: quantum money and quantum weak coin flipping. Quantum money exploits the no-cloning property of quantum physics to generate unforgeable tokens, banknotes, and credit cards. We provide the first proof-of-principle implementation of this task, using photonic systems at telecom wavelengths. We then develop a practical security proof for quantum credit card schemes, in which the bank can remotely verify a card even in the presence of a malicious payment terminal. We finally propose a setup for secure quantum storage of the credit card, using electromagnetically-induced transparency in a cloud of cold cesium atoms. Quantum weak coin flipping is a fundamental cryptographic primitive, which helps construct more complex tasks such as bit commitment and multiparty computation. It allows two distant parties to flip a coin when they both desire opposite outcomes. Using quantum entanglement then prevents any party from biasing the outcome of the flip beyond a certain probability. We propose the first implementation for quantum weak coin flipping, which requires a single photon and linear optics only. We provide the complete security analysis in the presence of noise and losses, and show that the protocol is implementable on the scale of a small city with current technology. We finally propose a linear-optical extension of the protocol to lower the coin bias.
• On the Learnability of Software Router Performance via CPU Measurements
  
  • Shelbourne Charles
  • Linguaglossa Leonardo
  • Lipani Aldo
  • Zhang Tianzhu
  • Geyer Fabien
  , 2019, pp.23-25. In the last decade the ICT community observed a growing popularity of software networking paradigms. This trend consists in moving network applications from static, expensive, hardware equipment (e.g. router, switches, firewalls) towards flexible, cheap pieces of software that are executed on a commodity server. In this context, a server owner may provide the server resources (CPUs, NICs, RAM) for customers, following a Service-Level Agreement (SLA) about clients' requirements. The problem of resource allocation is typically solved by overprovisioning, as the clients' application is opaque to the server owner, and the resource required by clients' applications are often unclear or very difficult to quantify. This paper shows a novel approach that exploits machine learning techniques in order to infer the input traffic load (i.e., the expected network traffic condition) by solely looking at the runtime CPU footprint. (10.1145/3360468.3366776)
  DOI : 10.1145/3360468.3366776
• Generalized Sliced Wasserstein Distances
  
  • Kolouri Soheil
  • Nadjahi Kimia
  • Şimşekli Umut
  • Badeau Roland
  • Rohde Gustavo K.
  , 2019.
• Comparing the Performance of State-of-the-Art Software Switches for NFV
  
  • Zhang Tianzhu
  • Linguaglossa Leonardo
  • Gallo Massimo
  • Giaccone Paolo
  • Iannone Luigi
  • Roberts James
  , 2019. Software switches are increasingly used in network function virtualization (NFV) to route traffic between virtualized network functions (VNFs) and physical network interface cards (NICs). Understanding of alternative switch designs remains deficient, however, in the absence of a comprehensive, comparative performance analysis. In this paper, we propose a methodology intended to be fair and use it to compare the performance of seven state-of-the-art software switches. We first explore their respective design spaces and then compare their performance under four representative test scenarios. Each scenario corresponds to a specific case of routing NFV traffic between NICs and/or VNFs. Our experimental results show that no single software switch prevails in all scenarios. It is therefore important to choose the one that is best adapted to a given use-case. The presented results and analysis bring a better understanding of design tradeoffs and identify potential bottlenecks that limit the performance of software switches.
• Real-Time Machine Learning Competition on Data Streams at the IEEE Big Data 2019
  
  • Boulegane Dihia
  • Radulovic Nedeljko
  • Bifet Albert
  • Fievet Ghislain
  • Sohn Jimin
  • Nam Yeonwoo
  • Yu Seojeong
  • Choi Dong-Wan
  , 2019, pp.3493-3497. (10.1109/BigData47090.2019.9006357)
  DOI : 10.1109/BigData47090.2019.9006357
• Weakly Supervised Representation Learning for Audio-Visual Scene Analysis
  
  • Parekh Sanjeel
  • Essid Slim
  • Ozerov Alexey
  • Duong Ngoc Q. K.
  • Pérez Patrick
  • Richard Gael
  IEEE/ACM Transactions on Audio, Speech and Language Processing, Institute of Electrical and Electronics Engineers, 2019. Audiovisual (AV) representation learning is an important task from the perspective of designing machines with the ability to understand complex events. To this end, we propose a novel multimodal framework that instantiates multiple instance learning. Specifically, we develop methods that identify events and localize corresponding AV cues in unconstrained videos. Importantly, this is done using weak labels where only video-level event labels are known without any information about their location in time. We show that the learnt representations are useful for performing several tasks such as event/object classification, audio event detection, audio source separation and visual object localization. An important feature of our method is its capacity to learn from unsynchronized audiovisual events. We also demonstrate our framework's ability to separate out the audio source of interest through a novel use of nonnegative matrix factorization. State-of-the-art classification results, with a F1-score of 65.0, are achieved on DCASE 2017 smart cars challenge data with promising generalization to diverse object types such as musical instruments. Visualizations of localized visual regions and audio segments substantiate our system's efficacy, especially when dealing with noisy situations where modality-specific cues appear asynchronously.
• On the Capacity of Block Fading Optical Wireless Channels
  
  • Li Longguang
  • Moser Stefan M
  • Wang Ligong
  • Wigger Michèle
  , 2019. This paper investigates the capacity of block fading optical intensity channels with more transmit than receive antennas under different assumptions on the transmitter's channel state information (CSI). Lower and upper bounds on the capacities are derived using the entropy power inequality (EPI) and a dual expression for capacity. Our lower bounds for perfect and partial CSI utilize a transmit-antenna cooperation strategy based on minimum-energy signaling, which we proposed recently. For perfect CSI, this lower bound matches the upper bound asymptotically in the high signal-to-noise ratio (SNR) regime. For imperfect CSI, our lower bound is close to its perfect-CSI counterpart. (10.1109/GLOBECOM38437.2019.9013614)
  DOI : 10.1109/GLOBECOM38437.2019.9013614
• Offloading strategies for mobile terminals with energy harvesting capabilities
  
  • Fawaz Ibrahim
  , 2019. Aujourd’hui, les communications mobiles sans fil sont en pleine croissance, en raison du grand nombre d’appareils connectés, ce qui augmente considérablement la demande de gros volumes de données, nécessitant des calculs intensifs et entraînant une forte consommation d’énergie. Toutefois, cette expansion des services sans fil est encore entravée par les limitations des terminaux mobiles, en termes de capacité de traitement, de stockage et d’énergie. Récemment, le Mobile Edge Computing (MEC) et la récupération d’énergie (EH) ont été proposés comme des technologies prometteuses pour prolonger la durée de vie des batteries des appareils mobiles et améliorer leurs capacités de traitement. D’une part, le MEC permet de décharger les tâches de calcul des appareils mobiles vers une station de base voisine avec plus d’énergie et de ressources de traitement. D’autre part, l’EH exploite des sources d’énergie alternatives et renouvelables pour alimenter les appareils mobiles. Cependant, la nature stochastique des énergies renouvelables peut entraîner des pénuries d’énergie. Dans ce cas, les performances du système peuvent être dégradées en raison de la perte de paquets ou d’une latence intolérable. Afin de garantir la durabilité du système, des politiques de transmission efficaces sous les contraintes de l’EH sont nécessaires. Dans cette thèse, nous étudions la planification conjointe des ressources et le déchargement des calculs dans un système MEC monoutilisateur fonctionnant avec des dispositifs basés sur l’EH. La contribution principale de ce travail est l’introduction de la contrainte de délai stricte au lieu de la contrainte de délai moyenne pour satisfaire les besoins futurs des communications à faible latence et des applications critiques. Nous étudions trois configurations différentes. Dans la première configuration, nous considérons que le canal est parfaitement connu au niveau de l’émetteur (CSI parfait) et nous visons à minimiser la perte de paquets due à la violation du délai et au débordement de la mémoire tampon du dispositif. Le problème d’optimisation associé est modélisé comme un processus de décision de Markov et la politique optimale est donnée par des techniques de programmation dynamique. Nous montrons que la politique optimale est plus performante que les autres politiques en adaptant le nombre de paquets traités aux états du système. Dans la seconde configuration, nous considérons un scénario plus réaliste, où le canal n’est pas parfaitement connu à l’émetteur et il est acquis après une phase d’estimation. En fait, cette estimation peut être erronée entraînant une dégradation supplémentaire du taux de perte de paquets. Par conséquent, nous évaluons la politique optimale obtenue précédemment lorsque le CSI est imparfait et nous montrons qu’elle reste robuste par rapport à d’autres politiques. Enfin, nous examinons la configuration sans CSI au niveau de l’émetteur. Nous supposons donc qu’un CSI obsolète est seulement disponible et nous montrons que la politique optimale proposée peut encore atteindre de bonnes performances par rapport à d’autres politiques
• Network of Experts: Learning from Evolving Data Streams Through Network-Based Ensembles
  
  • Gomes Heitor Murilo
  • Bifet Albert
  • Fournier-Viger Philippe
  • Granatyr Jones
  • Read Jesse
  , 2019, pp.704-716. (10.1007/978-3-030-36708-4_58)
  DOI : 10.1007/978-3-030-36708-4_58
• Asymptotic Guarantees for Learning Generative Models with the Sliced-Wasserstein Distance
  
  • Nadjahi Kimia
  • Durmus Alain
  • Şimşekli Umut
  • Badeau Roland
  , 2019.