Publications

2018

• Learning with Noise-Contrastive Estimation: Easing training by learning to scale
  
  • Labeau Matthieu
  • Allauzen Alexandre
  , 2018, pp.3090-3101. Noise-Contrastive Estimation (NCE) is a learning criterion that is regularly used to train neural language models in place of Maximum Likelihood Estimation, since it avoids the computational bottleneck caused by the output softmax. In this paper, we analyse and explain some of the weaknesses of this objective function, linked to the mechanism of self-normalization, by closely monitoring comparative experiments. We then explore several remedies and modifications to propose tractable and efficient NCE training strategies. In particular, we propose to make the scaling factor a trainable parameter of the model, and to use the noise distribution to initialize the output bias. These solutions, yet simple, yield stable and competitive performances in either small and large scale language modelling tasks.
• Hierarchical Graph Clustering using Node Pair Sampling
  
  • Bonald Thomas
  • Charpentier Bertrand
  • Galland Alexis
  • Hollocou Alexandre
  , 2018. We present a novel hierarchical graph clustering algorithm inspired by modularity-based clustering techniques. The algorithm is agglomerative and based on a simple distance between clusters induced by the probability of sampling node pairs. We prove that this distance is reducible, which enables the use of the nearest-neighbor chain to speed up the agglomeration. The output of the algorithm is a regular dendrogram, which reveals the multi-scale structure of the graph. The results are illustrated on both synthetic and real datasets.
• Novel compact transmitters for short-reach optical communication
  
  • Peucheret Christophe
  • Chaibi Mohamed E
  • Bramerie Laurent
  • Gay Mathilde
  • Hassan Karim
  • Erasme Didier
  • Perez-Galacho Diego
  • Marris-Morini Delphine
  , 2018. In this talk, we will review our recent work on the demonstration of novel compact transmitter structures employing Si or III-V integration for short reach optical communications. A special focus will be given on means to address the limitations induced by group-velocity dispersion for such systems. Customized modulators operating in the O-band, or employing single-sideband modulation in conjunction with single- (PAM2 and PAM4) or multi-carrier (OFDM) modulation in the C-band will be presented.
• Explicit formulas for photon number discrimination with on/off detectors
  
  • Miatto Filippo M
  • Safari Akbar
  • Boyd Robert
  Applied optics, Optical Society of America, 2018, 57 (23), pp.6750. Discriminating between Fock states with a high degree of accuracy is a desirable feature for modern applications of optical quantum information processing. A well-known alternative to sophisticated photon number discriminating detectors is to split the field among a number of simple on/off detectors and infer the desired quantity from the measurement results. In this work we find an explicit analytical expression of the detection probability for any number of input photons, any number of on/off detectors, and we include quantum efficiency and a false count probability. This allows us to explicitly invert the conditional probability using Bayes' theorem and express the number of photons that we had at the input in the most unbiased way possible with ready-to-use formulas. We conclude with some examples. (10.1364/AO.57.006750)
  DOI : 10.1364/AO.57.006750
• A Reproducible MEG/EEG Group Study With the MNE Software: Recommendations, Quality Assessments, and Good Practices
  
  • Jas Mainak
  • Larson Eric ​
  • Engemann Denis
  • Leppäkangas Jaakko
  • Taulu Samu
  • Hämäläinen Matti
  • Gramfort Alexandre
  Frontiers in Neuroscience, Frontiers, 2018, 12. Cognitive neuroscience questions are commonly tested with experiments that involve a cohort of subjects. The cohort can consist of a handful of subjects for small studies to hundreds or thousands of subjects in open datasets. While there exist various online resources to get started with the analysis of magnetoencephalography (MEG) or electroencephalography (EEG) data, such educational materials are usually restricted to the analysis of a single subject. This is in part because data from larger group studies are harder to share, but also analyses of such data often require subject-specific decisions which are hard to document. This work presents the results obtained by the reanalysis of an open dataset from Wakeman and Henson (2015) using the MNE software package. The analysis covers preprocessing steps, quality assurance steps, sensor space analysis of evoked responses, source localization, and statistics in both sensor and source space. Results with possible alternative strategies are presented and discussed at different stages such as the use of high-pass filtering versus baseline correction, tSSS vs. SSS, the use of a minimum norm inverse vs. LCMV beamformer, and the use of univariate or multivariate statistics. This aims to provide a comparative study of different stages of M/EEG analysis pipeline on the same dataset, with open access to all of the scripts necessary to reproduce this analysis. (10.3389/fnins.2018.00530)
  DOI : 10.3389/fnins.2018.00530
• Design, Synthesis and Application of A Novel Approximate Adder
  
  • Ban Tian
  • Wang Baokun
  • Naviner Lirida
  , 2018, pp.488-491. Approximate computing is a new design paradigm in VLSI design and test. It can improve the performance of error-tolerant applications at the expense of slight loss in computational accuracy. In this paper, we propose a novel approximate adder with a hybrid structure (HYB-adder) which produces results of different precision. The proposed adder is synthesized by utilizing 28nm FD-SOI (fully-depleted silicon-on-insulator) technology. The proposed HYB-adder outperforms existing approximate adder designs regarding mean error distance with comparable area, delay and power consumption. The efficiency is also validated by its application in DCT/IDCT procedures. (10.1109/MWSCAS.2018.8624023)
  DOI : 10.1109/MWSCAS.2018.8624023
• Applying source separation to music
  
  • Pardo Bryan
  • Liutkus Antoine
  • Duan Zhiyao
  • Richard Gael
  , 2018, Chapter 16. Separation of existing audio into remixable elements is very useful to repurpose music audio. Applications include upmixing video soundtracks to surround sound (e.g. home theater 5.1 systems), facilitating music transcriptions, allowing better mashups and remixes for disk jockeys, and rebalancing sound levels on multiple instruments or voices recorded simultaneously to a single track. In this chapter, we provide an overview of the algorithms and approaches designed to address the challenges and opportunities in music. Where applicable, we also introduce commonalities and links to source separation for video soundtracks, since many musical scenarios involve video soundtracks (e.g. YouTube recordings of live concerts, movie sound tracks). While space prohibits describing every method in detail, we include detail on representative music‐specific algorithms and approaches not covered in other chapters. The intent is to give the reader a high‐level understanding of the workings of key exemplars of the source separation approaches applied in this domain. (10.1002/9781119279860.ch16)
  DOI : 10.1002/9781119279860.ch16
• Hierarchical Self-awareness and Authority for Scalable Self-integrating Systems
  
  • Diaconescu Ada
  • Porter Barry
  • Rodrigues Roberto
  • Pournaras Evangelos
  , 2018.
• Ideologically-embedded design
  
  • Detienne Françoise
  • Baker Michael
  • Le Bail Chloé
  , 2018.
• Denoising of Microscopy Images: A Review of the State-of-the-Art, and a New Sparsity-Based Method
  
  • Meiniel William
  • Olivo-Marin Jean-Christophe
  • Angelini Elsa
  IEEE Transactions on Image Processing, Institute of Electrical and Electronics Engineers, 2018, 27 (8), pp.3842-3856. This paper reviews the state-of-the-art in denoising methods for biological microscopy images and introduces a new and original sparsity-based algorithm. The proposed method combines total variation (TV) spatial regularization, enhancement of low-frequency information, and aggregation of sparse estimators and is able to handle simple and complex types of noise (Gaussian, Poisson, and mixed), without any a priori model and with a single set of parameter values. An extended comparison is also presented, that evaluates the denoising performance of the thirteen (including ours) state-of-the-art denoising methods specifically designed to handle the different types of noises found in bioimaging. Quantitative and qualitative results on synthetic and real images show that the proposed method outperforms the other ones on the majority of the tested scenarios. (10.1109/TIP.2018.2819821)
  DOI : 10.1109/TIP.2018.2819821
• Telemetry-based stream-learning of BGP anomalies
  
  • Putina Andrian
  • Nivaggioli Patrice
  • Precup Cristina
  • Pletcher Drew
  • Bifet Albert
  • Rossi Dario
  , 2018.
• Enhancing data protection with a structure-wise fragmentation and dispersal of encrypted data
  
  • Kapusta Katarzyna
  • Memmi Gérard
  , 2018.
• Cooperative Anti-Jamming Relaying for Control Channel Jamming in Vehicular Networks
  
  • Gu Pengwenlong
  • Hua Cunqing
  • Khatoun Rida
  • Wu Yue
  • Serhrouchni Ahmed
  IEEE Transactions on Vehicular Technology, Institute of Electrical and Electronics Engineers, 2018, 67 (8), pp.7033-7046.
• Fair dropping for multi-resource fairness in software routers Extended Abstract
  
  • Addanki Vamsi
  • Linguaglossa Leonardo
  • Roberts James
  • Rossi Dario
  , 2018.
• ProvSQL: Provenance and Probability Management in PostgreSQL
  
  • Senellart Pierre
  • Jachiet Louis
  • Maniu Silviu
  • Ramusat Yann
  Proceedings of the VLDB Endowment (PVLDB), VLDB Endowment, 2018, 11 (12), pp.2034-2037. This demonstration showcases ProvSQL, an open-source module for the PostgreSQL database management system that adds support for computation of provenance and probabilities of query results. A large range of provenance formalisms are supported, including all those captured by provenance semirings, provenance semirings with monus, as well as where-provenance. Probabilistic query evaluation is made possible through the use of knowledge compilation tools, in addition to standard approaches such as enumeration of possible worlds and Monte-Carlo sampling. ProvSQL supports a large subset of non-aggregate SQL queries. (10.14778/3229863.3236253)
  DOI : 10.14778/3229863.3236253
• Unsupervised detection of ruptures in spatial relationships in video sequences based on log-likelihood ratio
  
  • Abou-Elailah Abdalbassir
  • Bloch Isabelle
  • Gouet-Brunet Valérie
  Pattern Analysis and Applications, Springer Verlag, 2018, 21 (3), pp.829-846. In this work, we propose a new approach to automatically detect ruptures in spatial relationships in video sequences, based on low-level primitives, in unsupervised manner. The spatial relationships between two objects of interest are modeled using angle and distance histograms as examples. The evolution of the spatial relationships during time is estimated from the distances between two successive angle or distance histograms and then considered as a temporal signal. The evolution of a spatial relationship is modeled by a linear Gaussian model. Then, two hypotheses "without change" and "with change" are considered, and a log-likelihood ratio is computed. The distribution of the log-likelihood ratio, given that H 0 is true, is estimated and used to compute the p value. The comparison of this p value to a significance level , expressing the probability of false alarms, allows us to detect significant ruptures in spatial relationships during time. In addition, this approach is generalized to detect multiple object events such as merging, splitting, and other events that contain ruptures in their spatial relationships evolution. This work shows that the description of spatial relationships across time is a promising step toward event detection. (10.1007/s10044-017-0669-9)
  DOI : 10.1007/s10044-017-0669-9
• VALIDATION OF A NONINVASIVE SYSTEM TO OBSERVE GLOTTAL OPENING AND CLOSING: EXTERNAL PHOTOGLOTTOGRAPH (ePGG)
  
  • Amelot Angelique
  • Sathiyanarayanan Darshan
  • Maeda Shinji
  • Honda Kyioshi
  • Crevier-Buchman Lise
  , 2018.
• Rényi Entropy Power Inequalities via Normal Transport and Rotation
  
  • Rioul Olivier
  Entropy, MDPI, 2018, 20 (9), pp.641. Following a recent proof of Shannon’s entropy power inequality (EPI), a comprehensive framework for deriving various EPIs for the Rényi entropy is presented that uses transport arguments from normal densities and a change of variable by rotation. Simple arguments are given to recover the previously known Rényi EPIs and derive new ones, by unifying a multiplicative form with constant c and a modification with exponent α of previous works. In particular, for log-concave densities, we obtain a simple transportation proof of a sharp varentropy bound.
• Spatio-Temporal Constrained Tone Mapping Operator for HDR Video Compression
  
  • Ozcinar Cagri
  • Lauga Paul
  • Valenzise Giuseppe
  • Dufaux Frédéric
  Journal of Visual Communication and Image Representation, Elsevier, 2018, 55, pp.166–178. With the growing popularity of high dynamic range (HDR) imaging, efficient compression techniques are demanded, as HDR video entails typically higher raw data rate than traditional video. For this purpose, we introduce a hybrid spatially and temporally constrained content-adaptive tone mapping operator (TMO) to convert the input HDR video into a tone mapped video sequence, which is then encoded using the high efficiency video coding (HEVC) standard. The proposed TMO simultaneously exploits intra-frame spatial redundancies and preserves inter-frame temporal coherence of the tone mapped video sequence. Extensive experimental results show that the developed spatio-temporal TMO (ST-TMO) solution yields higher coding performance than existing frame-by-frame TMO’s, and compares favorably with state-of-the-art methods based on a fixed transfer function. (10.1016/j.jvcir.2018.06.003)
  DOI : 10.1016/j.jvcir.2018.06.003
• Recall of mobile phone usage and laterality in young people: The multinational Mobi-Expo study
  
  • Goedhart Geertje
  • van Wel Luuk
  • Langer Chelsea
  • de Llobet Viladoms Patricia
  • Wiart Joe
  • Hours Martine
  • Kromhout Hans
  • Benke Geza
  • Bouka Evdoxia
  • Bruchim Revital
  • Choi Kyung-Hwa
  • Eng Amanda
  • Ha Mina
  • Huss Anke
  • Kiyohara Kosuke
  • Kojimahara Noriko
  • Krewski Daniel
  • Lacour Brigitte
  • ‘t Mannetje Andrea
  • Maule Milena
  • Migliore Enrica
  • Mohipp Charmaine
  • Momoli Franco
  • Petridou Eleni Th.
  • Radon Katja
  • Rémen Thomas
  • Sadetzki Siegal
  • Sim Malcolm
  • Weinmann Tobias
  • Cardis Elisabeth
  • Vrijheid Martine
  • Vermeulen Roel
  Environmental Research, Elsevier, 2018, 165, pp.150-157. (10.1016/j.envres.2018.04.018)
  DOI : 10.1016/j.envres.2018.04.018
• A hierarchical Bayesian perspective on majorization-minimization for non-convex sparse regression: application to M/EEG source imaging
  
  • Bekhti Yousra
  • Lucka Felix
  • Salmon Joseph
  • Gramfort Alexandre
  Inverse Problems, IOP Publishing, 2018, 34 (8), pp.085010. Majorization-minimization (MM) is a standard iterative optimization technique which consists in minimizing a sequence of convex surrogate functionals. MM approaches have been particularly successful to tackle inverse problems and statistical machine learning problems where the regularization term is a sparsity-promoting concave function. However, due to non-convexity, the solution found by MM depends on its initialization. Uniform initialization is the most natural and often employed strategy as it boils down to penalizing all coefficients equally in the first MM iteration. Yet, this arbitrary choice can lead to unsatisfactory results in severely under-determined inverse problems such as source imaging with magneto- and electro-encephalography (M/EEG). The framework of hierarchical Bayesian modeling (HBM) is an alternative approach to encode sparsity. This work shows that for certain hierarchical models, a simple alternating scheme to compute fully Bayesian maximum a posteriori(MAP) estimates leads to the exact same sequence of updates as a standard MM strategy (see the adaptive lasso). With this parallel outlined, we show how to improve upon these MM techniques by probing the multimodal posterior density using Markov Chain
• A practical method for measuring Web above-the-fold time
  
  • N. da Hora Diego
  • Christophides Vassilis
  • Teixeira Renata
  • Rossi Dario
  , 2018.
• A pratical method for measuring web above the fold time
  
  • da Hora Diego
  • Christophides Vassilis
  • Teixeira Renata
  • Rossi Dario
  , 2018.
• Accurate Characterization of Dynamic Cell Load in Noise-Limited Random Cellular Networks
  
  • Ghatak Gourab
  • de Domenico Antonio
  • Coupechoux Marceau
  , 2018, pp.1-5. <p>The analyses of cellular network performance based on stochastic geometry generally ignore the traffic dynamics in the network. This restricts the proper evaluation and di- mensioning of the network from the perspective of a mobile operator. To address the effect of dynamic traffic, recently, the mean cell approach has been introduced, which approximates the average network load by the zero cell load. However, this is not a realistic characterization of the network load, since a zero cell is statistically larger than a random cell drawn from the population of cells, i.e., a typical cell. In this paper, we analyze the load of a noise-limited network characterized by high signal to noise ratio (SNR). The noise-limited assumption can be applied to a variety of scenarios, e.g., millimeter wave networks with efficient interference management mechanisms. First, we provide an analytical framework to obtain the cumulative density function of the load of the typical cell. Then, we obtain two approximations of the average load of the typical cell. We show that our study provides a more realistic characterization of the average load of the network as compared to the mean cell approach. Moreover, the prescribed closed form approximation is more tractable than the mean cell approach.</p>
• Two Routes to Automata Minimization and the Ways to Reach It Efficiently
  
  • Lombardy Sylvain
  • Sakarovitch Jacques
  , 2018, 10977, pp.248-260. This paper reports on the work done for the implementation of the algorithms for the computation of the minimal quotient of an automaton in the Awali platform. In the case of non-deterministic or of weighted automata, the minimal quotient of an automaton is obtained by merging all states in bisimulation. Two strategies are explored for the computation of the coarsest bisimulation equivalence. The first one is an extension of the Moore algorithm for the computation of the minimal quotient of a DFA; the second one is inspired by the Hopcroft algorithm for the same problem. These two strategies yield algorithms with the same quadratic complexity and we study the cases where the second strategy can be improved in order to achieve a complexity similar to the one of Hopcroft algorithm. (10.1007/978-3-319-94812-6_21)
  DOI : 10.1007/978-3-319-94812-6_21