Publications

2019

• SILP: A Stochastic Imitative Learning Protocol for Multi-Carrier Spectrum Access
  
  • Iellamo Stefano
  • Coupechoux Marceau
  • Khan Zaheer
  IEEE Transactions on Cognitive Communications and Networking, IEEE, 2019, 5 (4), pp.990-1003. Decentralized wireless networks require efficient channel access protocols to enable wireless nodes (WNs) to access dedicated frequency channels without any coordination. In this paper, we develop a distributed spectrum access protocol for the case where the WNs are equipped with multiple radio transceivers. We consider the case where the channels are identical and duly separated so that each of the users' antenna can access only one of the available channels. To model the competition amongst WNs, we formulate a particular multi-agent multi-carrier spectrum access game, where each WN has to decide at each iteration how many antennas and which frequency channels it has to access. To study the resulting equilibrium, we solve a multi-objective optimization problem and design a bi-level learning algorithm which is proven to converge towards a socially efficient and max-min fair equilibrium state. (10.1109/TCCN.2019.2924925)
  DOI : 10.1109/TCCN.2019.2924925
• Measuring the Shattering coefficient of Decision Tree models
  
  • de Mello Rodrigo
  • Manapragada Chaitanya
  • Bifet Albert
  Expert Systems with Applications, Elsevier, 2019, 137, pp.443-452. (10.1016/j.eswa.2019.07.012)
  DOI : 10.1016/j.eswa.2019.07.012
• TRADI: Tracking deep neural network weight distributions
  
  • Franchi Gianni
  • Bursuc Andrei
  • Aldea Emmanuel
  • Dubuisson Séverine
  • Bloch Isabelle
  , 2019.
• Delay-optimal resource scheduling of energy harvesting based devices
  
  • Fawaz Ibrahim
  • Sarkiss Mireille
  • Ciblat Philippe
  IEEE Transactions on Green Communications and Networking, IEEE, 2019, 6 (4). This paper investigates resource scheduling in a wireless communication system operating with Energy Harvesting (EH) based devices and perfect Channel State Information (CSI). The aim is to minimize the packet loss that occurs when the buffer is overflowed or when the queued packet is older than a certain pre-defined threshold. We so consider a strict delay constraint rather than an average delay constraint. The associated optimization problem is modeled as Markov Decision Process (MDP) where the actions are the number of packets sent on the known channel at each slot. The optimal deterministic offline policy is exhibited through dynamic programming techniques, i.e. Value Iteration (VI) algorithm. We show that the gain in the number of transmitted packets and the consumed energy is substantial compared to: i) a naive policy which forces the system to send the maximum number of packets using the available energy in the battery, ii) two variants of the previous policy that take into account the buffer state, and iii) a policy optimized with an average delay constraint. Finally, we evaluate our optimal policy under imperfect CSI scenario where only an estimate of the channel state is available. (10.1109/TGCN.2019.2924242)
  DOI : 10.1109/TGCN.2019.2924242
• Some (almost) optimally extendable linear codes
  
  • Mesnager Sihem
  • Zhang Fengrong
  • Tang Chunming
  • Zhou Yong
  Designs, Codes and Cryptography, Springer Verlag, 2019, 87 (12), pp.2813-2834. (10.1007/s10623-019-00639-4)
  DOI : 10.1007/s10623-019-00639-4
• The Influence of CSI in Ultra-Reliable Low-Latency Communications with IR-HARQ
  
  • Avranas Apostolos
  • Kountouris Marios
  • Ciblat Philippe
  , 2019. Emerging 5G networks will need to efficiently support ultra-reliable, low-latency communication (URLLC), which requires extremely low latency (at msec order) with very high reliability (99.999%). In this work, we consider a URLLC system with incremental redundancy hybrid automatic repeat request (IR-HARQ) and investigate the effect of channel state information (CSI) at the transmitter on throughput and energy consumption optimization. For that, we analyze the feasibility region and the performance in block fading channels for the cases of full and statistical CSI. Our results show that the full CSI scheme is less robust and we also reveal a desirable balance between the trade-off quantities of energy and throughput.
• Exhaustive single bit fault analysis. A use case against Mbedtls and OpenSSL’s protection on ARM and Intel CPU
  
  • Carré Sébastien
  • Desjardins Matthieu
  • Facon Adrien
  • Guilley Sylvain
  Microprocessors and Microsystems: Embedded Hardware Design, Elsevier, 2019, 71, pp.102860. (10.1016/j.micpro.2019.102860)
  DOI : 10.1016/j.micpro.2019.102860
• Work-conserving dynamic TDM-based memory arbitration for multi-criticality real-time systems
  
  • Hebbache Farouk
  , 2019. Multi-core architectures pose many challenges in real-time systems, which arise from contention between concurrent accesses to shared memory. Among the available memory arbitration policies, Time-Division Multiplexing (TDM) ensures a predictable behavior by bounding access latencies and guaranteeing bandwidth to tasks independently from the other tasks. To do so, TDM guarantees exclusive access to the shared memory in a fixed time window. TDM, however, provides a low resource utilization as it is non-work-conserving. Besides, it is very inefficient for resources having highly variable latencies, such as sharing the access to a DRAM memory. The constant length of a TDM slot is, hence, highly pessimistic and causes an underutilization of the memory. To address these limitations, we present dynamic arbitration schemes that are based on TDM. However, instead of arbitrating at the level of TDM slots, our approach operates at the granularity of clock cycles by exploiting slack time accumulated from preceding requests. This allows the arbiter to reorder memory requests, exploit the actual access latencies of requests, and thus improve memory utilization. We demonstrate that our policies are analyzable as they preserve the guarantees of TDM in the worst case, while our experiments show an improved memory utilization. (10.70675/71b4611az889cz4cbezb5d4za8cfefc3ed01)
  DOI : 10.70675/71b4611az889cz4cbezb5d4za8cfefc3ed01
• Dynamic and nonlinear properties of quantum dot lasers for photonic integrated circuits on silicon
  
  • Duan Jianan
  , 2019. Silicon photonics have been introduced to overcome low efficiency and high energy consumption of telecom links using twisted pairs or coaxial cables. This technology provides novel functionality and high performance for applications in high speed communication systems, short reach optical interconnects, and the deployment of optical links from chipto-chip, board-to-board or rack-to-rack (datacom). Silicon is known as a very efficient semiconductor material for waveguiding light in particular owing to the strong index contrast with silica. However, the indirect bandgap of silicon makes light emission from silicon inefficient, and other techniques such as wafer- or flipchip bonding must be investigated if light emission is to be realized. The drawbacks of such heterogeneous integration concentrate on the high cost and the limited scalability. Lasers heterogeneously integrated on silicon are also more sensitive to optical reflections originating from the transition between passive/active interfaces. The best way to overcome these drawbacks is to move on to direct epitaxial growth of IIIV materials on silicon for photonics integration. In this context, quantum dot lasers using semiconductor atoms as a gain medium are ideal because they enable smaller devices, amplification with large thermal stability and high tolerance to epitaxial defects. Ultra-low noise optical transmitters are required not only for the coherent systems but also for future chipscale atomic clocks and radar related applications because of the sensitivity to the frequency noise and intensity noise can strongly affect the bit error rates. To this end, the first part of the thesis reports an intrinsic spectral linewidth as low as 80 kHz and a relative intensity noise less than - 150 dB/Hz in InAs/InP quantum dot lasers. In particular, it is shown that a small vertical coupling is more suitable for low intensity noise operation due to the suppression of the carrier noise in the excited state. The second part of the thesis investigates the dynamic and nonlinear properties of epitaxial quantum dot lasers on silicon. As mentioned above, lasers heterogeneously integrated on silicon are more sensitive to parasitic reflections. When combined with external optical feedback, the laser stability can be dramatically affected. As no on-chip optical isolators integrated with lasers and having sufficient isolation ratio exist, the development of feedback insensitive transmitters remains a major objective. This thesis presents an error-free transmission of an epitaxial quantum dot laser on silicon externally modulated at 10 Gb/s and subjected to 100% optical feedback. Such remarkable feedback insensitivity directly results from the near-zero linewidth enhancement factor, the large damping factor, the strong contrast between the ground state and excited states and a shorter carrier lifetime. These results pave the way for future high-performance photonics integrated circuits on silicon operating without optical isolators. (10.70675/9b0f4528z04fez419dz8603zcc7d1751fd3e)
  DOI : 10.70675/9b0f4528z04fez419dz8603zcc7d1751fd3e
• Exploration of multivariate EEG /MEG signals using non-stationary models
  
  • Ablin Pierre
  , 2019. Independent Component Analysis (ICA) models a set of signals as linear combinations of independent sources. This analysis method plays a key role in electroencephalography (EEG) and magnetoencephalography (MEG) signal processing. Applied on such signals, it allows to isolate interesting brain sources, locate them, and separate them from artifacts. ICA belongs to the toolbox of many neuroscientists, and is a part of the processing pipeline of many research articles. Yet, the most widely used algorithms date back to the 90's. They are often quite slow, and stick to the standard ICA model, without more advanced features.The goal of this thesis is to develop practical ICA algorithms to help neuroscientists. We follow two axes. The first one is that of speed. We consider the optimization problems solved by two of the most widely used ICA algorithms by practitioners: Infomax and FastICA. We develop a novel technique based on preconditioning the L-BFGS algorithm with Hessian approximation. The resulting algorithm, Picard, is tailored for real data applications, where the independence assumption is never entirely true. On M/EEG data, it converges faster than the `historical' implementations.Another possibility to accelerate ICA is to use incremental methods, which process a few samples at a time instead of the whole dataset. Such methods have gained huge interest in the last years due to their ability to scale well to very large datasets. We propose an incremental algorithm for ICA, with important descent guarantees. As a consequence, the proposed algorithm is simple to use and does not have a critical and hard to tune parameter like a learning rate.In a second axis, we propose to incorporate noise in the ICA model. Such a model is notoriously hard to fit under the standard non-Gaussian hypothesis of ICA, and would render estimation extremely long. Instead, we rely on a spectral diversity assumption, which leads to a practical algorithm, SMICA. The noise model opens the door to new possibilities, like finer estimation of the sources, and use of ICA as a statistically sound dimension reduction technique. Thorough experiments on M/EEG datasets demonstrate the usefulness of this approach.All algorithms developed in this thesis are open-sourced and available online. The Picard algorithm is included in the largest M/EEG processing Python library, MNE and Matlab library, EEGlab. (10.70675/85c09bc9z1441z4032zab16zbbddb31aea5b)
  DOI : 10.70675/85c09bc9z1441z4032zab16zbbddb31aea5b
• Statistical models for comprehensive meta-analysis of neuroimaging studies
  
  • Dockes Jérôme
  , 2019. Thousands of neuroimaging studies are published every year. Exploiting this huge amount of results is difficult. Indeed, individual studies lack statistical power and report many spurious findings. Even genuine effects are often specific to particular experimental settings and difficult to reproduce. Meta- analysis aggregates studies to identify consistent trends in reported associations between brain structure and behavior. The standard approach to meta-analysis starts by gathering a sample of studies that investigate a same mental process or disease.Then, a statistical test delineates brain regions where there is a significant agreement among reported findings. In this thesis, we develop a different kind of metaanalysis that focuses on prediction rather than hypothesis testing. We build predictive models that map textual descriptions of experiments, mental processes or diseases to anatomical regions in the brain. Our supervised learning approach comes with a natural quantitative evaluation framework, and we conduct extensive experiments to validate and compare statistical models. We collect and share the largest existing dataset of neuroimaging studies and stereotactic coordinates. This dataset contains the full text and locations of neurological observations for over 13 000 publications. In the last part, we turn to decoding: inferring mental states from brain activity.We perform this task through meta-analysis of fMRI statistical maps collected from an online data repository. We use fMRI data to distinguish a wide range of mental conditions. Standard meta-analysis is an essential tool to distinguish true discoveries from noise and artifacts. This thesis introduces methods for predictive metaanalysis, which complement the standard approach and help interpret neuroimaging results and formulate hypotheses or formal statistical priors. (10.70675/5d6e3b8dzcfd6z4d66za3b2zbd664cc598cd)
  DOI : 10.70675/5d6e3b8dzcfd6z4d66za3b2zbd664cc598cd
• Classification of Lightweight Block Ciphers for Specific Processor Accelerated Implementations
  
  • Tehrani Etienne
  • Graba Tarik
  • Si Merabet Abdelmalek
  • Guilley Sylvain
  • Danger Jean-Luc
  , 2019, pp.747-750. (10.1109/ICECS46596.2019.8965156)
  DOI : 10.1109/ICECS46596.2019.8965156
• Machine learning for streaming data
  
  • Gomes Heitor Murilo
  • Read Jesse
  • Bifet Albert
  • Barddal Jean Paul
  • Gama João
  SIGKDD explorations : newsletter of the Special Interest Group (SIG) on Knowledge Discovery & Data Mining, Association for Computing Machinery (ACM), 2019, 21 (2), pp.6-22. (10.1145/3373464.3373470)
  DOI : 10.1145/3373464.3373470
• Structures composites absorbantes radar pour applications navales (Conception, simulation, réalisation et mesures)
  
  • Parneix Patrick
  • Barka André
  • Rance Olivier
  • Soiron Michel
  • Lepage Anne Claire
  • Begaud Xavier
  , 2019.
• 3D Modeling for Propagation of UHF-RFID Tags’ Signals in an Indoor Environment
  
  • Hatem Elias
  • Abou-Chakra Sara
  • Colin Elizabeth
  • El-Hassan Bachar
  • Laheurte Jean-Marc
  , 2019, pp.1-6. (10.1109/MENACOMM46666.2019.8988525)
  DOI : 10.1109/MENACOMM46666.2019.8988525
• Prédiction de modèles structurés d'opinion : aspects théoriques et méthodologiques
  
  • Garcia Alexandre
  , 2019. Opinion mining has emerged as a hot topic in the machine learning community due to the recent availability of large amounts of opinionated data expressing customer's attitude towards merchandisable goods. Yet, predicting opinions is not easy due to the lack of computational models able to capture the complexity of the underlying objects at hand. Current approaches consist in predicting simple representations of the affective expressions, for example by restricting themselves to the valence attribute. This thesis focuses on the question of building structured output models able to jointly predict the different components of opinions in order to take advantage of the dependency between their parts. In this context, the choice of an opinion model has some consequences on the complexity of the learning problem and the statistical properties of the resulting predictors. We study 2 classical problems of opinion mining in which we instantiate squared surrogate based structured output learning techniques to illustrate the accuracy-complexity tradeoff arising when building opinion predictors. A second aspect of this thesis is to handle a newly released multimodal dataset containing entity and valence annotations at different granularity levels providing a complex representation of the underlying expressed opinions. We propose a deep learning based approach able to take advantage of the different labeled parts of the output objects by learning to jointly predict them. We propose a novel hierarchical architecture composed of different state-of-the-art multimodal neural layers and study the effect of different learning strategies in this joint prediction context. The resulting model is shown to improve over the performance of separate opinion component predictors and raises new questions concerning the optimal treatment of hierarchical labels in a structured prediction context. This thesis focuses on the question of building structured output models able to jointly predict the different components of opinions in order to take advantage of the dependency between their parts. In this context, the choice of an opinion model has some consequences on the complexity of the learning problem and the statistical properties of the resulting predictors. We specifically analyzed the case of preference based learning and joint entity and valence detection under a 2 layer binary tree representation in order to derive excess risk bounds and an analysis of the learning procedure algorithmic complexity. In these two settings, the output objects can be decomposed over a set of interacting parts with radical differences. However, we treat both problems under the same angle of squared surrogate based structured output learning and discuss the specificities of the two problem specifications. A second aspect of this thesis is to handle a newly released multimodal dataset containing entity and valence annotations at different granularity levels providing a complex representation of the underlying expressed opinions. In this context of large scale multimodal data with multiple granularity annotations, designing a dedicated model is quite challenging. Hence, we propose a deep learning based approach able to take advantage of the different labeled parts of the output objects by learning to jointly predict them. We propose a novel hierarchical architecture composed of different state-of-the-art multimodal neural layers and study the effect of different learning strategies in this joint prediction context. The resulting model is shown to improve over the performance of separate opinion component predictors and raises new questions concerning the optimal treatment of hierarchical labels in a structured prediction context. (10.70675/8e948d11z5644z4313z990bz0003f822852b)
  DOI : 10.70675/8e948d11z5644z4313z990bz0003f822852b
• Functional Isolation Forest
  
  • Staerman Guillaume
  • Mozharovskyi Pavlo
  • Clémençon Stéphan
  • d'Alché-Buc Florence
  , 2019. For the purpose of monitoring the behavior of complex infrastructures (e.g. aircrafts, transport or energy networks), high-rate sensors are deployed to capture multivariate data, generally unlabeled, in quasi continuous-time to detect quickly the occurrence of anomalies that may jeopardize the smooth operation of the system of interest. The statistical analysis of such massive data of functional nature raises many challenging methodological questions. The primary goal of this paper is to extend the popular Isolation Forest (IF) approach to Anomaly Detection, originally dedicated to finite dimensional observations, to functional data. The major difficulty lies in the wide variety of topological structures that may equip a space of functions and the great variety of patterns that may characterize abnormal curves. We address the issue of (randomly) splitting the functional space in a flexible manner in order to isolate progressively any trajectory from the others, a key ingredient to the efficiency of the algorithm. Beyond a detailed description of the algorithm, computational complexity and stability issues are investigated at length. From the scoring function measuring the degree of abnormality of an observation provided by the proposed variant of the IF algorithm, a Functional Statistical Depth function is defined and discussed as well as a multivariate functional extension. Numerical experiments provide strong empirical evidence of the accuracy of the extension proposed.
• Evaluating Datalog via Tree Automata and Cycluits
  
  • Amarilli Antoine
  • Bourhis Pierre
  • Monet Mikaël
  • Senellart Pierre
  Theory of Computing Systems, Springer Verlag, 2019, 63 (7), pp.1620-1678. We investigate parameterizations of both database instances and queries that make query evaluation fixed-parameter tractable in combined complexity. We show that clique-frontier-guarded Datalog with stratified negation (CFG-Datalog) enjoys bilinear-time evaluation on structures of bounded treewidth for programs of bounded rule size. Such programs capture in particular conjunctive queries with simplicial decompositions of bounded width, guarded negation fragment queries of bounded CQ-rank, or two-way regular path queries. Our result is shown by translating to alternating two-way automata, whose semantics is defined via cyclic provenance circuits (cycluits) that can be tractably evaluated. (10.1007/s00224-018-9901-2)
  DOI : 10.1007/s00224-018-9901-2
• Impact des technologies quantiques sur les communications et le traitement de l'information
  
  • Alléaume Romain
  , 2019.
• Towards a Reliable Machine Learning Based Global Misbehavior Detection in C-ITS: Model Evaluation Approach
  
  • Mahmoudi Issam
  • Kamel Joseph
  • Ben-Jemaa Ines
  • Kaiser Arnaud
  • Urien Pascal
  , 2019. Global misbehavior detection in Cooperative Intelligent Transport Systems (C-ITS) is carried out by a central entity named Misbe-havior Authority (MA). The detection is based on local misbehavior detection information sent by Vehicle's On-Board Units (OBUs) and by RoadSide Units (RSUs) called Misbehavior Reports (MBRs) to the MA. By analyzing these Misbehavior Reports (MBRs), the MA is able to compute various misbehavior detection information. In this work, we propose and evaluate different Machine Learning (ML) based solutions for the internal detection process of the MA. We show through extensive simulation and several detection metrics the ability of solutions to precisely identify different misbehavior types.
• Fiberered tunable heralded single photon source at telecom wavelength
  
  • Cordier Martin
  • Benabid Fetah
  • Delaye Philippe
  • Zaquine Isabelle
  , 2019.
• Vers une conception de systèmes réactifs synchrones sûrs
  
  • Chabane Sarah
  • Ameur-Boulifa Rabéa
  • Mohamed Mezghiche
  , 2019. La nécessité de gérer la complexité croissante des systèmes en général, et les systèmes embarqués en particulier pousse leur conception vers une approche basée sur la réutilisation de composants existants.À cette complexité s'ajoute des exigences techniques, les systèmes doivent satisfaire des contraintes strictes de fiabilité et de correction. Dans notre travail, nous préconisons d'offrir un cadre théorique pour le développement de composants réactifs synchrones sûrs de manière compositionnelle. Dans ce papier nous offrons d'une part, un cadre de description de composants réactifs synchronesélémentaires dans un formalisme adapté pour la vérification formelle de propriétés de sûreté, et d'autre part, un cadre pour la construction des systèmes globauxà partir de composantsélémentaires par une opération de composition garantissant la correction par construction.
• Multi-scale computational rhythm analysis : a framework for sections, downbeats, beats, and microtiming
  
  • Fuentes Magdalena
  , 2019. Computational rhythm analysis deals with extracting and processing meaningful rhythmical information from musical audio. It proves to be a highly complex task, since dealing with real audio recordings requires the ability to handle its acoustic and semantic complexity at multiple levels of representation. Existing methods for rhythmic analysis typically focus on one of those levels, failing to exploit music’s rich structure and compromising the musical consistency of automatic estimations. In this work, we propose novel approaches for leveraging multi-scale information for computational rhythm analysis. Our models account for interrelated dependencies that musical audio naturally conveys, allowing the interplay between different time scales and accounting for music coherence across them. In particular, we conduct a systematic analysis of downbeat tracking systems, leading to convolutional-recurrent architectures that exploit short and long term acoustic modeling; we introduce a skip-chain conditional random field model for downbeat tracking designed to take advantage of music structure information (i.e. music sections repetitions) in a unified framework; and we propose a language model for joint tracking of beats and micro-timing in Afro-Latin American music. Our methods are systematically evaluated on a diverse group of datasets, ranging from Western music to more culturally specific genres, and compared to state-of-the-art systems and simpler variations. The overall results show that our models for downbeat tracking perform on par with the state of the art, while being more musically consistent. Moreover, our model for the joint estimation of beats and microtiming takes further steps towards more interpretable systems. The methods presented here offer novel and more holistic alternatives for computational rhythm analysis, towards a more comprehensive automatic analysis of music. (10.70675/edf67d07z82f1z4dc5z8084zbd6f15906b38)
  DOI : 10.70675/edf67d07z82f1z4dc5z8084zbd6f15906b38
• Unitary $t$-designs from $relaxed$ seeds
  
  • Mezher Rawad
  • Ghalbouni Joe
  • Dgheim Joseph
  • Markham Damian
  , 2019. In this work we reduce the requirements for generating $t$-designs, an important tool for randomisation with applications across quantum information and physics. We show that random quantum circuits with support over families of $relaxed$ finite sets of unitaries which are approximately universal in $U(4)$ (we call such sets $seeds$), converge towards approximate unitary $t$-designs efficiently in $poly(n,t)$ depth, where $n$ is the number of inputs of the random quantum circuit, and $t$ is the order of the design. We show this convergence for seeds which are relaxed in the sense that every unitary matrix in the seed need not have an inverse in the seed, nor be composed entirely of algebraic entries in general, two requirements which have restricited previous constructions. We suspect the result found here is not optimal, and can be improved. Particularly because the number of gates in the relaxed seeds introduced here grows with $n$ and $t$. We conjecture that constant sized seeds such as those in (Brand\~ao, Harrow, and Horodecki, Commun. Math. Phys. 2016) are sufficient.
• Run or Hide? Both! A Method Based on IPv6 Address Switching to Escape While Being Hidden
  
  • Ayrault Maxime
  • Borde Etienne
  • Kühne Ulrich
  , 2019, pp.47-56. (10.1145/3338468.3356827)
  DOI : 10.1145/3338468.3356827