Publications

2018

• Dual-rotation C-arm cone-beam tomographic acquisition and reconstruction frameworks for low-contrast detection in brain soft-tissue imaging
  
  • Reshef Aymeric
  , 2018. Interventional C-arm systems are real-time X-ray imaging systems, that can perform tomographic acquisitions by rotating the C-arm around the patient ; however, C-arm cone-beam computed tomography (CBCT) achieves a lower contrast resolution than diagnostic CT, which is necessary in order to benefit from the clinical information of soft tissues in the brain. We propose a new C-arm CBCT acquisition and reconstruction framework to increase low-contrast detection in brain soft-tissue imaging. In order to emulate a bow-tie filter, a dualrotation acquisition is proposed. To account for all the specificities of the dual-rotation acquisition, a dedicated iterative reconstruction algorithm is designed, that includes the ramp filter in the cost function. By switching from filtered backprojection (FBP) to backprojection-filtration (BPF) reconstruction methods, we propose an alternative, direct reconstruction method for dual-rotation acquisitions. For single-rotation acquisitions, the method ensures to perform as good as FBP with arbitrarily coarse angular sampling in planar geometries, and provides a different approximation from the Feldkamp-Davis-Kress (FDK) algorithm in the cone-beam geometry. Although we used it to emulate a virtual bow-tie, our dual-rotation acquisition framework is intrinsically related to region-of-interest (ROI) imaging through the truncated acquisition. With few or no modification of the proposed reconstruction methods, we successfully addressed the problem of ROI imaging in the context of dual-rotation acquisitions.
• Ferrous Alloys Quality from Antiquity to Middle Age: A Statistical and Diachronic Approach
  
  • Gosselin Manon
  • Tendero Yohann
  • Arribet-Deroin Danielle
  • Téreygeol Florian
  • Pagès Gaspard
  • Dillmann Philippe
  , 2018.
• MAIN MELODY EXTRACTION WITH SOURCE-FILTER NMF AND CRNN
  
  • Basaran Dogac
  • Essid Slim
  • Peeters Geoffroy
  , 2018. Estimating the main melody of a polyphonic audio recording remains a challenging task. We approach the task from a classification perspective and adopt a convolutional recurrent neural network (CRNN) architecture that relies on a particular form of pretraining by source-filter nonneg-ative matrix factorisation (NMF). The source-filter NMF decomposition is chosen for its ability to capture the pitch and timbre content of the leading voice/instrument, providing a better initial pitch salience than standard time-frequency representations. Starting from such a musically motivated representation, we propose to further enhance the NMF-based salience representations with CNN layers , then to model the temporal structure by an RNN network and to estimate the dominant melody with a final classification layer. The results show that such a system achieves state-of-the-art performance on the MedleyDB dataset without any augmentation methods or large training sets.
• Progressive and Efficient Multi-Resolution Representations for Brain Tractograms
  
  • Mercier Corentin
  • Gori Pietro
  • Rohmer D.
  • Cani Marie-Paule
  • Boubekeur T
  • Thiery Jean-Marc
  • Bloch Isabelle
  , 2018. Current tractography methods generate tractograms composed of millions of 3D polylines, called fibers, making visualization and interpretation extremely challenging, thus complexifying the use of this technique in a clinical environment. We propose to progressively simplify tractograms by grouping similar fibers into generalized cylinders. This produces a fine-grained multi-resolution model that provides a progressive and real-time navigation through different levels of detail. This model preserves the overall structure of the tractogram and can be adapted to different measures of similarity. We also provide an efficient implementation of the method based on a Delaunay tetrahedralization. We illustrate our method using the Human Connectome Project dataset.
• A Spectral Algorithm with Additive Clustering for the Recovery of Overlapping Communities in Networks
  
  • Kaufmann Emilie
  • Bonald Thomas
  • Lelarge Marc
  Theoretical Computer Science, Elsevier, 2018, 742, pp.3-26. This paper presents a novel spectral algorithm with additive clustering designed to identify overlapping communities in networks. The algorithm is based on geometric properties of the spectrum of the expected adjacency matrix in a random graph model that we call stochastic blockmodel with overlap (SBMO). An adaptive version of the algorithm, that does not require the knowledge of the number of hidden communities, is proved to be consistent under the SBMO when the degrees in the graph are (slightly more than) logarithmic. The algorithm is shown to perform well on simulated data and on real-world graphs with known overlapping communities. (10.1016/j.tcs.2017.12.028)
  DOI : 10.1016/j.tcs.2017.12.028
• Arabic Cyberbullying Detection: Using Deep Learning
  
  • Haidar Batoul
  • Chamoun Maroun
  • Serhrouchni Ahmed
  , 2018, pp.284-289. As much as internet and smart devices are taking a big role in the lives of children and adolescents, also the threat of Cyberbullying on the lives and wellbeing of those youngsters is rising. The threat of cyberbullying is acknowledged around the world generally and in the Arabic areas specifically. A lot of research is done for finding automated solutions for cyberbullying detection in several languages, but not much has been done for Arabic Language. At the other hand, a lot of interest is invested in Deep Learning techniques, where Deep Learning has been applied in several areas and showed vast success. Thus this paper proposes a solution that employs Deep Learning methods in the process of Arabic Cyberbullying Detection. Specifically a Feed Forward Neural Network is trained on an Arabic Dataset for the purpose of cyberbullying detection. (10.1109/ICCCE.2018.8539303)
  DOI : 10.1109/ICCCE.2018.8539303
• A deep learning architecture to detect events in EEG signals during sleep
  
  • Chambon Stanislas
  • Thorey Valentin
  • Arnal Pierrick J
  • Mignot Emmanuel
  • Gramfort Alexandre
  , 2018. Electroencephalography (EEG) during sleep is used by clinicians to evaluate various neurological disorders. In sleep medicine, it is relevant to detect macro-events (≥ 10 s) such as sleep stages, and micro-events (≤ 2 s) such as spindles and K-complexes. Annotations of such events require a trained sleep expert, a time consuming and tedious process with a large inter-scorer variability. Automatic algorithms have been developed to detect various types of events but these are event-specific. We propose a deep learning method that jointly predicts locations, durations and types of events in EEG time series. It relies on a convolutional neural network that builds a feature representation from raw EEG signals. Numerical experiments demonstrate efficiency of this new approach on various event detection tasks compared to current state-of-the-art, event specific, algorithms.
• Sécurisation de l'Internet des objets
  
  • Hammi Mohamed Tahar
  , 2018. L'Internet des Objets ou en anglais the Internet of Things (IoT) représente aujourd'hui une partie majeure de notre vie quotidienne. Des milliards d'objets intelligents et autonomes, à travers le monde sont connectés et communiquent entre eux. Ce paradigme révolutionnaire crée une nouvelle dimension qui enlèveles frontières entre le monde réel et le monde virtuel. Son succès est dû à l’évolution des équipements matériels et des technologies de communication notamment sans fil. L’IoT est le fruit du développement et de la combinaison de différentes technologies. Il englobe presque tous les domaines de la technologie d’information (Information Technology (IT)) actuels.Les réseaux de capteurs sans fil représentent une pièce maîtresse du succès de l'IoT. Car en utilisant des petits objets qui sont généralement limités en terme de capacité de calcul, de mémorisation et en énergie, des environnements industriels, médicaux, agricoles, et autres peuvent être couverts et gérés automatiquement.La grande puissance de l’IoT repose sur le fait que ses objets communiquent, analysent, traitent et gèrent des données d’une manière autonome et sans aucune intervention humaine. Cependant, les problèmes liés à la sécurité freinent considérablement l’évolution et le déploiement rapide de cette haute echnologie. L'usurpation d’identité, le vols d’information et la modification des données représentent un vrai danger pour ce système des systèmes.Le sujet de ma thèse consiste en la création d'un système de sécurité permettant d’assurer les services d’authentification des objets connectés, d’intégrité des données échangées entres ces derniers et de confidentialité des informations. Cette approche doit prendre en considération les contraintes des objets et des technologies de communication utilisées.
• Query Answering with Transitive and Linear-Ordered Data
  
  • Amarilli Antoine
  • Benedikt Michael
  • Bourhis Pierre
  • Vanden Boom Michael
  Journal of Artificial Intelligence Research, Association for the Advancement of Artificial Intelligence, 2018, 63, pp.191-264. (10.1613/jair.1.11240)
  DOI : 10.1613/jair.1.11240
• Text to brain: predicting the spatial distribution of neuroimaging observations from text reports
  
  • Dockès Jérôme
  • Wassermann Demian
  • Poldrack Russell
  • Suchanek Fabian M.
  • Thirion Bertrand
  • Varoquaux Gaël
  , 2018, pp.1-18. Despite the digital nature of magnetic resonance imaging, the resulting observations are most frequently reported and stored in text documents. There is a trove of information untapped in medical health records, case reports, and medical publications. In this paper, we propose to mine brain medical publications to learn the spatial distribution associated with anatomical terms. The problem is formulated in terms of minimization of a risk on distributions which leads to a least-deviation cost function. An efficient algorithm in the dual then learns the mapping from documents to brain structures. Empirical results using coordinates extracted from the brain-imaging literature show that i) models must adapt to semantic variation in the terms used to describe a given anatomical structure, ii) voxel-wise parameterization leads to higher likelihood of locations reported in unseen documents, iii) least-deviation cost outperforms least-square. As a proof of concept for our method, we use our model of spatial distributions to predict the distribution of specific neurological conditions from text-only reports.
• AtlasNet: Multi-atlas Non-linear Deep Networks for Medical Image Segmentation
  
  • Vakalopoulou Maria
  • Chassagnon Guillaume
  • Bus Norbert
  • Marini Silva Rafael
  • Zacharaki Evangelia I.
  • Revel Marie-Pierre
  • Paragios Nikos
  , 2018. Deep learning methods have gained increasing attention in addressing segmentation problems for medical images analysis despite the challenges inherited from the medical domain, such as limited data availability, lack of consistent textural or salient patterns, and high di-mensionality of the data. In this paper, we introduce a novel multi-network architecture that exploits domain knowledge to address those challenges. The proposed architecture consists of multiple deep neural networks that are trained after co-aligning multiple anatomies through multi-metric deformable registration. This multi-network architecture can be trained with fewer examples and leads to better performance, robustness and generalization through consensus. Comparable to human accuracy, highly promising results on the challenging task of interstitial lung disease segmentation demonstrate the potential of our approach.
• Procédé sécurisé d’aide à une conduite fiable et sûre pour le franchissement d'un passage à niveau communicant
  
  • Monteuuis Jean-Philippe
  • Zhang Jun J.
  • Labiod Houda
  • Servel Alain
  • Mafrica Stefano
  , 2018.
• Attacker model for Connected and Automated Vehicles
  
  • Monteuuis Jean-Philippe
  • Labiod Houda
  • Zhang Jun J.
  • Servel Alain
  • Mafrica Stefano
  , 2018.
• Time warp invariant dictionary learning for time series clustering: application to music data stream analysis
  
  • Yazdi Saeed Varasteh
  • Douzal-Chouakria Ahlame
  • Gallinari Patrick
  • Moussallam Manuel
  , 2018. This work proposes a time warp invariant sparse coding and dictionary learning framework for time series clustering, where both input samples and atoms define time series of different lengths that involve variable delays. For that, first an l0 sparse coding problem is formalised and a time warp invariant orthogonal matching pursuit based on a new cosine maximisation time warp operator is proposed. A dictionary learning under time warp is then formalised and a gradient descent solution is developed. Lastly, a time series clustering based on the time warp sparse coding and dictionary learning is presented. The proposed approach is evaluated and compared to major alternative methods on several public datasets, with an application to deezer music data stream clustering.
• Varnishes of painting material studied by terahertz spectroscopy
  
  • Cassar Q.
  • Koch-Dandolo C.L.
  • Guillet J.P.
  • Roux M.
  • Fauquet F.
  • Mounaix Patrick
  , 2018, pp.1-2. (10.1109/IRMMW-THz.2018.8510065)
  DOI : 10.1109/IRMMW-THz.2018.8510065
• Multi-Packet HARQ with Delayed Feedback
  
  • Khreis Alaa
  • Ciblat Philippe
  • Bassi Francesca
  • Duhamel Pierre
  , 2018. In current wireless communication systems, the feedback required by the Hybrid Automatic ReQuest (HARQ) mechanism is received with some delay at the transmitter side. To alleviate this issue, parallel Stop-and-Wait HARQ is usually employed. In this paper, we propose a multi-packet HARQ protocol (also called superposition coding or multi-layer HARQ) to improve the user's delay distribution and increase the throughput, without any additional feedback such as Channel State Information. The performance analysis, provided from an information-theoretic point-of-view, shows that the proposed protocol offers better delay distribution, higher throughput and lower message error rate compared to the conventional parallel Stop-and-Wait HARQ, at the expense of increased decoding complexity. (10.1109/PIMRC.2018.8580804)
  DOI : 10.1109/PIMRC.2018.8580804
• Varnishes of painting material studied by terahertz spectroscopy
  
  • Cassar Q.
  • Koch-Dandolo C.L.
  • Guillet J.P.
  • Roux M.
  • Fauquet F.
  • Mounaix Patrick
  , 2018, pp.1-2. (10.1109/IRMMW-THz.2018.8510065)
  DOI : 10.1109/IRMMW-THz.2018.8510065
• Weakly Supervised Object Detection in Artworks
  
  • Gonthier Nicolas
  • Gousseau Yann
  • Ladjal Saïd
  • Bonfait Olivier
  , 2018. We propose a method for the weakly supervised detection of objects in paintings. At training time, only image-level annotations are needed. This, combined with the efficiency of our multiple-instance learning method, enables one to learn new classes on-the-fly from globally annotated databases, avoiding the tedious task of manually marking objects. We show on several databases that dropping the instance-level annotations only yields mild performance losses. We also introduce a new database, IconArt, on which we perform detection experiments on classes that could not be learned on photographs, such as Jesus Child or Saint Sebastian. To the best of our knowledge, these are the first experiments dealing with the automatic (and in our case weakly supervised) detection of iconographic elements in paintings. We believe that such a method is of great benefit for helping art historians to explore large digital databases. (10.1007/978-3-030-11012-3_53)
  DOI : 10.1007/978-3-030-11012-3_53
• Single Qubit Arbitrary Unitary Synthesis using Photonic Spectral Encoding
  
  • Raghunathan Ravi
  • Guillaume Ricard
  • Miatto Filippo M
  • Zaquine Isabelle
  • Alleaume Romain
  , 2018.
• Self-Improving System Integration - Status and Challenges after Five Years of SISSY
  
  • Bellman Kirstie
  • Botev Jean
  • Diaconescu Ada
  • Esterle Lukas
  • Gruhl Christian
  • Landauer Chris
  • Lewis Peter
  • Stein Anthony
  • Tomforde Sven
  • Wurtz Rolf
  , 2018, pp.160-167. The self-improving system integration (SISSY) initiative has emerged in recent years in response to a systems engineering trend towards the organisation of open, interconnected systems integrating a large set of heterogeneous and autonomous subsystems. Based on the idea to equip subsystems with capabilities to assess and maintain their own integration status within the overall system composition, a variety of concepts, techniques, and contributions have been proposed and fruitfully discussed at the particular events of the underlying workshop series. In this article, we summarise and categorise these research efforts and derive a roadmap towards full-scale SISSY systems. (10.1109/FAS-W.2018.00042)
  DOI : 10.1109/FAS-W.2018.00042
• Transport Mode Detection when Fine-grained and Coarse-grained Data Meet
  
  • Asgari Fereshteh
  • Clémençon Stéphan
  , 2018, pp.301-307. Transport Mode Detection (TDM) algorithms in principle are developed for fine-grained data which is either high frequent accurate GPS data with/or further optional data such as accelerometer from mobile phones. The main drawback of using high frequent GPS data is the battery issue which makes it very expensive experiment to be employed for large scale data. Besides, GPS can not cover underground trajectories and some additional resource is required for such multi-modal trajectories. In this work we investigate the TDM algorithms using a combination of fine-grained (GPS) and coarse-grained (GSM) data with lower frequency compared to existing studies. We first provide a comprehensive overview of transport mode detection for such data by exploring both segment based and sequence-based machine learning approaches and then we use the collected heterogeneous mobility dataset to compare different mode detection algorithms. With the obtained results, we show that TDM algorithms are still effective approach for noisy and sparse heterogeneous data. The obtained decent performance provides the opportunity of extracting precious data from a large population of users in an inexpensive approach.
• Signaux Optiques
  
  • Gallion Philippe
  , 2018.
• Unified Stochastic Reverberation Modeling
  
  • Badeau Roland
  , 2018. In the field of room acoustics, it is well known that reverberation can be characterized statistically in a particular region of the time-frequency domain (after the transition time and above Schroeder’s frequency). Since the 1950s, various formulas have been established, focusing on particular aspects of reverberation: exponential decay over time, correlations between frequencies, correlations between sensors at each frequency, and time-frequency distribution. In this paper, we introduce a new stochastic reverberation model, that permits us to retrieve all these well-known results within a common mathematical framework. To the best of our knowledge, this is the first time that such a unification work is presented. The benefits are multiple: several new formulas generalizing the classical results are established, that jointly characterize the spatial, temporal and spectral properties of late reverberation.
• Multi-task Feature Learning for EEG-based Emotion Recognition Using Group Nonnegative Matrix Factorization
  
  • Hajlaoui Ayoub
  • Chetouani Mohamed
  • Essid Slim
  , 2018, pp.91-95. (10.23919/EUSIPCO.2018.8553390)
  DOI : 10.23919/EUSIPCO.2018.8553390
• A Concept for Proactive Knowledge Construction in Self-Learning Autonomous Systems
  
  • Stein Anthony
  • Tomforde Sven
  • Diaconescu Ada
  • Hähner Jörg
  • Müller-Schloer Christian
  , 2018, pp.204-213. The research initiative of self-improving and self-integrating systems (SISSY) emerged as response to the dramatically increasing complexity in information and communication technology. Such systems' ability of autonomous online learning has been identified as a key enabler for SISSY as well as for the broader field of self-adaptive and self-organizing (SASO) systems, since it provides the technical basis for dealing with the inherent dynamics of non-stationary environments that continually challenge these systems with unforeseen situations, disturbances, and changing goals. However, the learning progress is guided by the experiences in terms of situations the system has been exposed to so far – this reactive learning strategy naturally results in missing or inappropriate knowledge. In this paper, we define a formal system model and formulate an abstract learning task for SISSY systems. We further introduce the notion of knowledge and knowledge gaps to subsequently present a novel concept to automatically assess a system's existing knowledge base and, consequently, to proactively acquire knowledge to prepare SISSY/SASO systems for coping with disturbances and other changes that occur at runtime. By the proposed a priori construction of knowledge, we pursue the overall goal to increase the robustness as well as the learning efficiency of self-learning autonomous systems. Endowing these systems with the ability of identifying regions in their knowledge base that are not appropriately covered, strengthens their self-awareness property. (10.1109/FAS-W.2018.00048)
  DOI : 10.1109/FAS-W.2018.00048