Publications

2018

• Normalizing constants of log-concave densities
  
  • Brosse Nicolas
  • Durmus Alain
  • Moulines Éric
  Electronic Journal of Statistics, Shaker Heights, OH : Institute of Mathematical Statistics, 2018, 12 (1), pp.851-889. We derive explicit bounds for the computation of normalizing constants Z for log-concave densities $\pi= \mathrm{e}^{−U} /Z$ w.r.t. the Lebesgue measure on $\mathbb{R}^d$. Our approach relies on a Gaussian annealing combined with recent and precise bounds on the Unadjusted Langevin Algorithm (Durmus, A. and Moulines, E. (2016). High-dimensional Bayesian inference via the Unadjusted Langevin Algorithm). Polynomial bounds in the dimension $d$ are obtained with an exponent that depends on the assumptions made on $U$. The algorithm also provides a theoretically grounded choice of the annealing sequence of variances. A numerical experiment supports our findings. Results of independent interest on the mean squared error of the empirical average of locally Lipschitz functions are established. (10.1214/18-EJS1411)
  DOI : 10.1214/18-EJS1411
• On subtrees of the representation tree in rational base numeration systems
  
  • Akiyama Shigeki
  • Marsault Victor
  • Sakarovitch Jacques
  Discrete Mathematics and Theoretical Computer Science, DMTCS, 2018, 20 (1), pp.1-33. Every rational number p/q defines a rational base numeration system in which every integer has a unique finite representation, up to leading zeroes. This work is a contribution to the study of the set of the representations of integers. This prefix-closed subset of the free monoid is naturally represented as a highly non-regular tree. Its nodes are the integers, its edges bear labels taken in {0,1,...,p-1}, and its subtrees are all distinct. We associate with each subtree (or with its root n) three infinite words. The bottom word of n is the lexicographically smallest word that is the label of a branch of the subtree. The top word of n is defined similarly. The span-word of n is the digitwise difference between the latter and the former. First, we show that the set of all the span-words is accepted by an infinite automaton whose underlying graph is essentially the same as the tree itself. Second, we study the function that computes for all n the bottom word associated with n+1 from the one associated with n, and show that it is realised by an infinite sequential transducer whose underlying graph is once again essentially the same as the tree itself. An infinite word may be interpreted as an expansion in base p/q after the radix point, hence evaluated to a real number. If T is a subtree whose root is n, then the evaluations of the labels of the branches of T form an interval of R . The length of this interval is called the span of n and is equal to the evaluation of the span-word of n. The set of all spans is then a subset of R and we use the preceding construction to study its topological closure. We show that it is an interval when p is greater than or equal to 2q-1, and a Cantor set of measure zero otherwise. (10.23638/DMTCS-20-1-10)
  DOI : 10.23638/DMTCS-20-1-10
• Streaming Data Mining with Massive Online Analytics (MOA)
  
  • Bifet Albert
  • Read Jesse
  • Holmes Geoff
  • Pfahringer Bernhard
  , 2018, 83, pp.1-25. Fast Big Data is being produced at high-velocity in real-time. To effectively deal with this type of streaming data produced in real time, we need to be able to adapt to changes on the distribution of the data being produced, and we need to do it using the minimum amount of time and memory. The Internet of Things (IoT) is a good example and motivation of this type of streaming data produced in real time. Massive Online Analytics (MOA) is a software environment for implementing algorithms and running experiments for online learning from evolving data streams. MOA is designed to deal with the challenging problem of scaling up the implementation of state of the art algorithms to real world dataset sizes. MOA includes classification and clustering methods. It contains collection of offline and online methods as well as tools for evaluation. MOA supports bi-directional interaction with WEKA, the Waikato Environment for Knowledge Analysis, and is released under the GNU GPL license. (10.1142/9789813228047_0001)
  DOI : 10.1142/9789813228047_0001
• Leveraging Inter-domain Stability for BGP Dynamics Analysis
  
  • Green Thomas
  • Lambert Anthony
  • Pelsser Cristel
  • Rossi Dario
  , 2018. In the Internet, Autonomous Systems continuously exchange routing information via the BGP protocol: the large number of networks involved and the verbosity of BGP result in a huge stream of updates. Making sense of all those messages remains a challenge today. In this paper, we leverage the notion of "primary path" (i.e., the most used inter-domain path of a BGP router toward a destination prefix for a given time period), reinterpreting updates by grouping them in terms of primary paths unavailability periods, and illustrate how BGP dynamics analysis would benefit from working with primary paths. Our contributions are as follows. First, through measurements, we validate the existence of primary paths: by analyzing BGP updates announced at the LINX RIS route collector spanning a three months period, we show that primary paths are consistently in use during the observation period. Second, we quantify the benefits of primary paths for BGP dynamics analysis on two use cases : Internet tomography and anomaly detection. For the latter, using three months of anomalous BGP events documented by BGPmon as reference, we show that primary paths could be used for detecting such events (hijacks and outages), testifying of the increased semantic they provide.
• Binaural Localization of Multiple Sound Sources by Non-Negative Tensor Factorization
  
  • Benaroya Laurent
  • Obin Nicolas
  • Liuni Marco
  • Roebel Axel
  • Raumel Wilson
  • Argentieri Sylvain
  IEEE/ACM Transactions on Audio, Speech and Language Processing, Institute of Electrical and Electronics Engineers, 2018, pp.1 - 1. —This paper presents non-negative factorization of audio signals for the binaural localization of multiple sound sources within realistic and unknown sound environments. Non-negative tensor factorization (NTF) provides a sparse representation of multi-channel audio signals in time, frequency, and space that can be exploited in computational audio scene analysis and robot audition for the separation and localization of sound sources. In the proposed formulation, each sound source is represented by mean of spectral dictionaries, temporal activation, and its distribution within each channel (here, left and right ears). This distribution, being dependent on the frequency, can be interpreted as an explicit estimation of the Head-Related Transfer Function (HRTF) of a binaural head which can then be converted into the estimated sound source position. Moreover, the semi-supervised formulation of the non-negative factorization allows to integrate prior knowledge about some sound sources of interest whose dictionaries can be learned in advance, whereas the remaining sources are considered as background sound which remains unknown and is estimated on-the-fly. The proposed NTF-based sound source localization is here applied to binaural sound source localization of multiple speakers within realistic sound environments. (10.1109/TASLP.2018.2806745)
  DOI : 10.1109/TASLP.2018.2806745
• A closer look at IP-ID behavior in the Wild
  
  • Salutari Flavia
  • Cicalese Danilo
  • Rossi Dario
  , 2018. Originally used to assist network-layer fragmentation and reassembly, the IP identification field (IP-ID) has been used and abused for a range of tasks, from counting hosts behind NAT, to detect router aliases and, lately, to assist detection of censorship in the Internet at large. These inferences have been possible since, in the past, the IP- ID was mostly implemented as a simple packet counter: however, this behavior has been discouraged for security reasons and other policies, such as random values, have been suggested. In this study, we propose a framework to classify the different IP-ID behaviors using active probing from a single host. Despite being only minimally intrusive, our technique is significantly accurate (99% true positive classification) robust against packet losses (up to 20%) and lightweight (few packets suffices to discriminate all IP-ID behaviors). We then apply our technique to an Internet-wide census, where we actively probe one alive target per each routable /24 subnet: we find that that the majority of hosts adopts a constant IP-IDs (39%) or local counter (34%), that the fraction of global counters (18%) significantly diminished, that a non marginal number of hosts have an odd behavior (7%) and that random IP-IDs are still an exception (2%).
• Comparison of Nonlinear Compensation Techniques for 400-Gb/s Coherent Multi-Band OFDM Super-Channels
  
  • Vgenopoulou Vasiliki
  • Song Mengdi
  • Pincemin Erwan
  • Jaouën Yves
  • Sygletos Stylianos
  • Roudas Ioannis
  Applied Sciences, Multidisciplinary digital publishing institute (MDPI), 2018, 8 (3), pp.447. (10.3390/app8030447)
  DOI : 10.3390/app8030447
• Buffer Management for Synchronous and Low-Latency Playback of Multi-Stream User-Generated Content
  
  • Potetsianakis Emmanouil
  • Le Feuvre J.
  , 2018. (10.1109/MIPR.2018.00022)
  DOI : 10.1109/MIPR.2018.00022
• The Conflicted Usage of RLUTs for Security-Critical Applications on FPGA
  
  • Basu Roy Debapriya
  • Bhasin Shivam
  • Danger Jean-Luc
  • Guilley Sylvain
  • He Wei
  • Mukhopadhyay Debdeep
  Journal of Hardware and Systems Security, 2018. (10.1007/s41635-018-0035-4)
  DOI : 10.1007/s41635-018-0035-4
• Energy Disaggregation for Commercial Buildings: A Statistical Analysis
  
  • Henriet Simon
  • Şimşekli Umut
  • Richard Gael
  • Fuentes Benoît
  , 2018.
• Lattice Codes for the Insertion and Deletion Channels
  
  • Solé Patrick
  • Sok L.
  • Tchamkerten Aslan
  • Belfiore Jean-Claude
  IEEE Transactions on Information Theory, Institute of Electrical and Electronics Engineers, 2018, 64 (3), pp.1595-1603.
• Probability Aware Fault-Injection Approach for SER Estimation
  
  • Armelin Fabio Batagin
  • Naviner Lirida
  • d'Amore Roberto
  , 2018.
• Optical Networks Throughput Enhancement via TCP Stop-and-Wait on Hybrid Switches
  
  • Minakhmetov Artur
  • Ware Cédric
  • Iannone Luigi
  , 2018. We report on possible 50% throughput increase in Optical Packet Switching (OPS) data-center networks by replacing all-optical switches with optical switches with shared electronic buffers further enhanced with TCP Stop-and-Wait algorithms.
• Real-time Downstream 25 Gbps PAM4 for High Speed TDM-PONs with both 25 and 12.5 Gbps ONUs
  
  • Barthomeuf Sylvain
  • Saliou Fabienne
  • Anet Neto Luiz
  • Le Guyader Bertrand
  • Chanclou Philippe
  • Erasme Didier
  , 2018. We experimentally achieved 40km, real-time, N1 ODN class (29dB) compliant PAM4 transmission with PAM4/NRZ receptions on the same PON. The emitted PAM4 is interpreted either as PAM4 or NRZ to allow full or half line-rates.
• Drone-borne GPR design: Propagation issues
  
  • Chandra Madhu
  • Tanzi Tullio
  Comptes Rendus. Physique, Académie des sciences (Paris), 2018, 19 (1-2), pp.72-84.
• Using FPGA self-produced transients to emulate SETs for SER estimation
  
  • Armelin Fabio Batagin
  • Naviner Lirida
  • d'Amore Roberto
  , 2018.
• Broadband 2-W Output Power Tandem Thulium-Doped Single Clad Fiber Amplifier at 2 μm
  
  • Tench Robert
  • Romano Clément
  • Delavaux Jean-Marc
  IEEE Photonics Technology Letters, Institute of Electrical and Electronics Engineers, 2018, 30 (5), pp.503-506.
• Improved Converses and Gap-Results for Coded Caching,
  
  • Wang Chien-Yi
  • Saeedi Bidokhti Shirin
  • Wigger Michèle
  IEEE Transactions on Information Theory, Institute of Electrical and Electronics Engineers, 2018, 64 (3), pp.1957-1976.
• Broadband 2-W Output Power Tandem Thulium-Doped Single Clad Fiber Amplifier at 2 $\mu$ m
  
  • Tench Robert E
  • Romano Clément
  • Delavaux Jean-Marc
  IEEE Photonics Technology Letters, Institute of Electrical and Electronics Engineers, 2018, 30 (5), pp.503 - 506. We report experimental and simulated performance of a tandem (dual-stage) Tm-doped silica fiber amplifier with a high signal output power of 2.6 W in the 2 µm band. Combined high dynamic range, high gain, low noise figure, and high OSNR are achieved with our design. (10.1109/LPT.2018.2801840)
  DOI : 10.1109/LPT.2018.2801840
• Imagerie satellitaire et IA : que font les réseaux de neurones profonds
  
  • Maître Henri
  , 2018. L'indexation des images satellitaires utilise aujourd'hui abondamment les méthodes classiques d'apprentissage automatique (détection de primitives, classification, etc.) mais pas les réseaux de neurones profonds malgré des besoins notables en sémantique. Pourquoi ?
• Sampling Constrained Asynchronous Communication: How to Sleep Efficiently
  
  • Chandar Venkat
  • Tchamkerten Aslan
  IEEE Transactions on Information Theory, Institute of Electrical and Electronics Engineers, 2018, 64 (3), pp.1867-1878. The minimum energy, and, more generally, the minimum cost, to transmit one bit of information was recently derived for bursty communication when information is available infrequently at random times at the transmitter. Furthermore, it was shown that even if the receiver is constrained to sample only a fraction ρ ∈ (0, 1] of the channel outputs, there is no capacity penalty. That is, for any strictly positive sampling rate ρ, the asynchronous capacity per unit cost is the same as under full sampling, i.e., when ρ = 1. Moreover, there is no penalty in terms of decoding delay. These results are asymptotic in nature, considering the limit as the number B of bits to be transmitted tends to infinity, while the sampling rate ρ remains fixed. A natural question is then whether the sampling rate ρ(B) can drop to zero without introducing a capacity (or delay) penalty compared with full sampling. We answer this question affirmatively. The main result of this paper is an essentially tight characterization of the minimum sampling rate. We show that any sampling rate that grows at least as fast as ω(1/B) is achievable, while any sampling rate smaller than o(1/B) yields unreliable communication. The key ingredient in our improved achievability result is a new, multi-phase adaptive sampling scheme for locating transient changes, which we believe may be of independent interest for certain change-point detection problems. (10.1109/TIT.2017.2771208)
  DOI : 10.1109/TIT.2017.2771208
• Autonomous system for data collection: from signal to information
  
  • Tanzi Tullio
  • Isnard Jean
  , 2018.
• Intelligence artificielle: défis scientifiques et attentes socio-économiques
  
  • Clémençon Stéphan
  Annales des Mines - Enjeux Numériques, Conseil général de l'Économie, ministère de l'Économie et des Finances, 2018 (1), pp.78-80. Au-delà de l’effervescence médiatique qui les entoure aujourd’hui, des craintes et des espoirs qu’ils suscitent, les succès récents de l’intelligence artificielle, et plus particulièrement ceux du machine-learning ‒ ce domaine à l’interface des disciplines mathématique et informatique qui vise à développer des techniques pour l’analyse automatique de données massives, à des fins prédictives le plus souvent ‒, reposent en grande majorité sur des concepts scientifiques formulés il y a maintenant plusieurs décennies, alors même qu’ils nous sont présentés comme les signes avant-coureurs d’une révolution qui n’épargnera aucun secteur de l’activité humaine.
• TRISK: A local features extraction framework for texture-plus-depth content matching
  
  • Karpushin Maxim
  • Valenzise Giuseppe
  • Dufaux Frederic
  Image and Vision Computing, Elsevier, 2018, 71, pp.1-16. In this paper we present a new complete detector–descriptor framework for local features extraction from grayscale texture-plus-depth images. It is designed by putting together a locally normalized binary descriptor and the popular AGAST corner detector modified to incorporate the depth map into the keypoint detection process. With these new local features, we target image matching applications when significant out-of-plane rotations and viewpoint position changes are present in the input data. Our approach is designed to perform on RGBD images acquired with low-cost sensors such as Kinect without any complex depth map preprocessing such as denoising or inpainting. We show improved results with respect to several other highly competitive local image features through both a classic local feature evaluation procedure and an illustrative application scenario. Moreover, the proposed method requires low computational effort. (10.1016/j.imavis.2017.11.007)
  DOI : 10.1016/j.imavis.2017.11.007
• Capacity Enhancement of Few-Mode Fiber Transmission Systems Impaired by Mode-Dependent Loss
  
  • Amhoud El-Mehdi
  • Rekaya-Ben Othman Ghaya
  • Jaouën Yves
  Applied Sciences, Multidisciplinary digital publishing institute (MDPI), 2018, 8 (3), pp.326. (10.3390/app8030326)
  DOI : 10.3390/app8030326