HAL : Dernières publications

• [hal-03145345] Learning elementary structures for 3D shape generation and matching

  We propose to represent shapes as the deformation and combination of learnable elementary 3D structures, which are primitives resulting from training over a collection of shape. We demonstrate that the learned elementary 3D structures lead to clear improvements in 3D shape generation and matching. More precisely, we present two complementary approaches for learning elementary structures: (i) patch deformation learning and (ii) point translation learning. Both approaches can be extended to abstract structures of higher dimensions for improved results. We evaluate our method on two tasks: reconstructing ShapeNet objects and estimating dense correspondences between human scans (FAUST inter challenge). We show 16% improvement over surface deformation approaches for shape reconstruction and outperform FAUST inter challenge state of the art by 6%.

• [hal-02104041] Discovering Visual Patterns in Art Collections with Spatially-consistent Feature Learning

  Our goal in this paper is to discover near duplicate patterns in large collections of artworks. This is harder than standard instance mining due to differences in the artistic media (oil, pastel, drawing, etc), and imperfections inherent in the copying process. The key technical insight is to adapt a standard deep feature to this task by fine-tuning it on the specific art collection using self-supervised learning. More specifically, spatial consistency between neighbouring feature matches is used as supervisory fine-tuning signal. The adapted feature leads to more accurate style-invariant matching, and can be used with a standard discovery approach, based on geometric verification, to identify duplicate patterns in the dataset. The approach is evaluated on several different datasets and shows surprisingly good qualitative discovery results. For quantitative evaluation of the method, we annotated 273 near duplicate details in a dataset of 1587 artworks attributed to Jan Brueghel and his workshop. Beyond artwork, we also demonstrate improvement on localization on the Oxford5K photo dataset as well as on historical photograph localization on the Large Time Lags Location (LTLL) dataset.

• [hal-03140696] Marceline Desbordes-Valmore sur Wikisource : plus accessible, donc plus enseignée ?

  De nombreux ouvrages de Marceline Desbordes-Valmore sont disponibles en version numérisée sur internet, par exemple sur Gallica, Google Books ou Numelyo. Pourtant, peu d'entre eux sont pleinement accessibles en version texte, permettant d'en copier/coller facilement des extraits. L'association Le deuxième texte y travaille actuellement, dans le cadre de ses activités pour mettre à disposition plus de textes écrits par des femmes, à destination notamment des enseignantes et enseignants de lettres, afin de diversifier les corpus littéraires enseignés du collège à l'université.

• [hal-03140604] Distributionally robust and risk-averse optimisation for the stochastic multi-product disassembly line balancing problem with workforce assignment

  Existing works usually focus on the single-product disassembly line balancing problem (DLBP). In practice, end-of-life (EOL) products to be disassembled may be heterogeneous, and the actual processing time of each task may vary with its assigned worker. This work studies a stochastic multi-product DLBP with workforce assignment, to minimise the system cost. Due to historical data scarcity, we assume that only partial distributional information of uncertain task processing times is known. Exceeding the preset cycle time may lead to a disassembly performance reduction, thus we control the cycle time violation via conditional Value-at-Risk (CVaR) constraints, i.e. in a risk-averse fashion. For the problem, we first propose a novel formulation with distributionally robust CVaR constraints. Then some valid inequalities are proposed, leading to an improved model. Two solution approaches, i.e. an exact cutting-plane method and an approximation method, are further proposed and compared, via numerical experiments. Some managerial insights are also drawn.

• [hal-03136103] Visualizing hierarchies in scRNA-seq data using a density tree-biased autoencoder

  Single cell RNA sequencing (scRNA-seq) data makes studying the development of cells possible at unparalleled resolution. Given that many cellular differentiation pro- cesses are hierarchical, their scRNA-seq data is expected to be approximately tree-shaped in gene expression space. Inference and representation of this tree-structure in two dimensions is highly desirable for biological interpretation and exploratory analysis. Our two contributions are an approach for identifying a meaningful tree structure from high-dimensional scRNA-seq data, and a visualization method respecting the tree-structure. We extract the tree structure by means of a density based minimum spanning tree on a vector quantization of the data and show that it captures biological information well. We then introduce DTAE, a tree-biased autoencoder that emphasizes the tree structure of the data in low dimensional space. We com- pare to other dimension reduction methods and demonstrate the success of our method experimentally. Our implementation relying on PyTorch and Higra is available at github.com/hci-unihd/DTAE.

• [tel-03127055] Reconstruction et correspondance de formes par apprentissage

  L'objectif de cette thèse est de développer des approches d'apprentissage profond pour modéliser et analyser les formes 3D. Les progrès dans ce domaine pourraient démocratiser la création artistique d'actifs 3D, actuellement coûteuse en temps et réservés aux experts du domaine. Nous nous concentrons en particulier sur deux tâches clefs pour la modélisation 3D : la reconstruction à vue unique et la mise en correspondance de formes.Une méthode de reconstruction à vue unique (SVR) prend comme entrée une seule image et prédit le monde physique qui a produit cette image. SVR remonte aux premiers jours de la vision par ordinateur. Étant donné que plusieurs configurations de formes, de textures et d'éclairage peuvent expliquer la même image il faut formuler des hypothèses sur la distribution d'images et de formes 3D pour résoudre l’ambiguïté. Dans cette thèse, nous apprenons ces hypothèses à partir de jeux de données à grande échelle au lieu de les concevoir manuellement. Les méthodes d'apprentissage nous permettent d'effectuer une reconstruction complète et réaliste de l'objet, y compris des parties qui ne sont pas visibles dans l'image d'entrée.La mise en correspondance de forme vise à établir des correspondances entre des objets 3D. Résoudre cette tâche nécessite à la fois une compréhension locale et globale des formes 3D qui est difficile à obtenir explicitement. Au lieu de cela, nous entraînons des réseaux neuronaux sur de grands jeux de données pour capturer ces connaissances implicitement.La mise en correspondance de forme a de nombreuses applications en modélisation 3D telles que le transfert d'attribut, le gréement automatique pour l'animation ou l'édition de maillage.La première contribution technique de cette thèse est une nouvelle représentation paramétrique des surfaces 3D modélisées par les réseaux neuronaux. Le choix de la représentation des données est un aspect critique de tout algorithme de reconstruction 3D. Jusqu'à récemment, la plupart des approches profondes en génération 3D prédisaient des grilles volumétriques de voxel ou des nuages de points, qui sont des représentations discrètes. Au lieu de cela, nous présentons une approche qui prédit une déformation paramétrique de surface, c'est-à-dire une déformation d'un modèle source vers une forme objectif. Pour démontrer les avantages ses avantages, nous utilisons notre nouvelle représentation pour la reconstruction à vue unique. Notre approche, baptisée AtlasNet, est la première approche profonde de reconstruction à vue unique capable de reconstruire des maillages à partir d'images sans s’appuyer sur un post-traitement indépendant, et peut le faire à une résolution arbitraire sans problèmes de mémoire. Une analyse plus détaillée d’AtlasNet révèle qu'il généralise également mieux que les autres approches aux catégories sur lesquelles il n'a pas été entraîné.Notre deuxième contribution est une nouvelle approche de correspondance de forme purement basée sur la reconstruction par des déformations. Nous montrons que la qualité des reconstructions de forme est essentielle pour obtenir de bonnes correspondances, et donc introduisons une optimisation au moment de l'inférence pour affiner les déformations apprises. Pour les humains et d'autres catégories de formes déformables déviant par une quasi-isométrie, notre approche peut tirer parti d'un modèle et d'une régularisation isométrique des déformations. Comme les catégories présentant des variations non isométriques, telles que les chaises, n'ont pas de modèle clair, nous apprenons à déformer n'importe quelle forme en n'importe quelle autre et tirons parti des contraintes de cohérence du cycle pour apprendre des correspondances qui respectent la sémantique des objets. Notre approche de correspondance de forme fonctionne directement sur les nuages de points, est robuste à de nombreux types de perturbations, et surpasse l'état de l'art de 15% sur des scans d'humains réels

• [hal-03134003] Synchronizing Almost-Group Automata

  In this paper we address the question of synchronizing random automata in the critical settings of almost-group automata. Group automata are automata where all letters act as permutations on the set of states, and they are not synchronizing (unless they have one state). In almost-group automata, one of the letters acts as a permutation on n−1 states, and the others as permutations. We prove that this small change is enough for automata to become synchronizing with high probability. More precisely, we establish that the probability that a strongly connected almost-group automaton is not synchronizing is (2<sup>k−1</sup>−1)/(n<sup>2(k−1)</sup>)*(1+o(1)), for a k-letter alphabet.

• [hal-03134002] Birecurrent sets

  A set is called recurrent if its minimal automaton is strongly connected and birecurrent if it is recurrent as well as reversal. We prove a series of results concerning birecurrent sets. It is already known that any birecurrent set is completely reducible (that is, such that the minimal representation of its characteristic series is completely reducible). The main result of this paper characterizes completely reducible sets as linear combinations of birecurrent sets

• [hal-03133246] Robust Angular Local Descriptor Learning

  [...]

• [hal-03132460] Label Propagation Guided by Hierarchy of Partitions for Superpixel Computation

  [...]

• [hal-03132263] Geometry-Consistent Generative Adversarial Networks for One-Sided Unsupervised Domain Mapping

  Unsupervised domain mapping aims to learn a function G XY to translate domain X to Y in the absence of paired examples. Finding the optimal G XY without paired data is an ill-posed problem, so appropriate constraints are required to obtain reasonable solutions. While some prominent constraints such as cycle consistency and distance preservation successfully constrain the solution space, they overlook the special properties of images that simple geometric transformations do not change the image's semantic structure. Based on this special property, we develop a geometry-consistent generative adversarial network (Gc-GAN), which enables one-sided unsupervised domain mapping. GcGAN takes the original image and its counterpart image transformed by a predefined geometric transformation as inputs and generates two images in the new domain coupled with the corresponding geometry-consistency constraint. The geometry-consistency constraint reduces the space of possible solutions while keep the correct solutions in the search space. Quantitative and qualitative comparisons with the baseline (GAN alone) and the state-of-the-art methods including CycleGAN [66] and Dis-tanceGAN [5] demonstrate the effectiveness of our method.

• [hal-03132245] Occlusion Boundary: A Formal Definition & Its Detection via Deep Exploration of Context

  Occlusion boundaries contain rich perceptual information about the underlying scene structure and provide important cues in many visual perception-related tasks such as object recognition, segmentation, motion estimation, scene understanding, and autonomous navigation. However, there is no formal definition of occlusion boundaries in the literature, and state-of-the-art occlusion boundary detection is still suboptimal. With this in mind, in this paper we propose a formal definition of occlusion boundaries for related studies. Further, based on a novel idea, we develop two concrete approaches with different characteristics to detect occlusion boundaries in video sequences via enhanced exploration of contextual information (e.g., local structural boundary patterns, observations from surrounding regions, and temporal context) with deep models and conditional random fields. Experimental evaluations of our methods on two challenging occlusion boundary benchmarks (CMU and VSB100) demonstrate that our detectors significantly outperform the current state-of-the-art. Finally, we empirically assess the roles of several important components of the proposed detectors to validate the rationale behind these approaches.

• [hal-03132172] Compacted binary trees admit a stretched exponential

  [...]

• [hal-03132149] Phase transitions from $\exp(n^{1/2})$ to $\exp(n^{2/3})$ in the asymptotics of banded plane partitions

  [...]

• [hal-03131027] OrthoNet: Multilayer Network Data Clustering

  [...]

• [hal-03128939] An O(n²) algorithm for time-bound adjustments for the cumulative scheduling problem

  [...]

• [hal-03034652] Efficient and Lightweight Polynomial-based Key Management Scheme for Dynamic Networks

  Wireless sensor networks and Internet of Things (IoT) are part of dynamic networks as new nodes can join while existing members can leave the system at any time. These networks mainly suffer from severe resource constraints like energy, storage and computation, which makes securing communications between nodes a real challenge. Several key establishment protocols have been proposed in the literature. Some of them are based on symmetric polynomials. However, the latter solutions have some limitations, such as the resilience to node capture attacks as well as the storage and computation overheads that are high for constrained nodes. In this paper, we propose a lightweight polynomial-based key management scheme for dynamic networks. The proposed scheme allows nodes to be able to establish secure communications between them, and ensures dynamism by supporting node addition and deletion after the setup phase. It also resists to node capture attack. The performance evaluation shows that our scheme reduces both the storage and computation overheads when compared to other related polynomial-based protocols.

• [hal-02564349] Convergence of constant step stochastic gradient descent for non-smooth non-convex functions

  This paper studies the asymptotic behavior of the constant step Stochastic Gradient Descent for the minimization of an unknown function F , defined as the expectation of a non convex, non smooth, locally Lipschitz random function. As the gradient may not exist, it is replaced by a certain operator: a reasonable choice is to use an element of the Clarke subdifferential of the random function; an other choice is the output of the celebrated backpropagation algorithm, which is popular amongst practionners, and whose properties have recently been studied by Bolte and Pauwels [7]. Since the expectation of the chosen operator is not in general an element of the Clarke subdifferential BF of the mean function, it has been assumed in the literature that an oracle of BF is available. As a first result, it is shown in this paper that such an oracle is not needed for almost all initialization points of the algorithm. Next, in the small step size regime, it is shown that the interpolated trajectory of the algorithm converges in probability (in the compact convergence sense) towards the set of solutions of the differential inclusion. Finally, viewing the iterates as a Markov chain whose transition kernel is indexed by the step size, it is shown that the invariant distribution of the kernel converge weakly to the set of invariant distribution of this differential inclusion as the step size tends to zero. These results show that when the step size is small, with large probability, the iterates eventually lie in a neighborhood of the critical points of the mean function F .

• [hal-03121428] Quasi-regularity verification for 2D polygonal objects based on medial axis analysis

  Quasi-regularity has been proved to be a sufficient condition for simple-connectedness preservation during the digitization process of 2D continuous objects. The original definition of quasi-regularity relies on set-based morphological operations of erosion and dilation. With this definition, quasi-regularity is algorithmically difficult to assess. In this paper, we propose a tractable framework for quasi-regularity verification, especially designed for polygons. Our approach mainly relies on the computation and analysis of the medial axis of these objects, and determines their potential quasi-regularity, and thus their ability to undergo a digitization without alteration of their topological properties. The framework is applied in the context of topology-preserving rigid motions of digital objects.

• [hal-03116273] Preface : Fixed Points in Computer Science (FICS) 2013

  This special issue of Fundemanta Informaticae contains revised and extended versions of contributions that were presented at the 6th edition of the workshop Fixed Points in Computer Science (FICS) 2013. The FICS 2013 workshop is was held on September 1st 2013 in Torino, as a satellite event ofthe EACSL Annual Conference on Computer Science Logic (CSL) 2013. Fixed points play a fundamental role in several areas of computer science. They are used tojustify (co)recursive definitions and associated reasoning techniques. The construction and properties of fixed points have been investigated in many different settings such as: design and implementationof programming languages, logics, verification, databases. The FICS workshop series aims to providea forum for researchers of the computer science and logic communities who study or apply the theoryof fixed points.The special issue presents 7 papers.

• [hal-03115362] Component Tree Loss Function: Definition and Optimization

  In this article, we propose a method to design loss functions based on component trees which can be optimized by gradient descent algorithms and which are therefore usable in conjunction with recent machine learning approaches such as neural networks. We show how the altitudes associated to the nodes of such hierarchical image representations can be differentiated with respect to the image pixel values. This feature is used to design a generic loss function that can select or discard image maxima based on various attributes such as extinction values. The possibilities of the proposed method are demonstrated on simulated and real image filtering.

• [hal-03113554] Ultrametric fitting by gradient descent

  [...]

• [hal-03112500] Thinning Algorithms Classification

  [...]

• [hal-03112278] Subcritical Random Hypergraphs, High-Order Components, and Hypertrees

  [...]

• [hal-01745165] Concurrent computation of topological watershed on shared memory parallel machines

  The watershed transform is considered as the most appropriate method for image seg-mentation in the field of mathematical morphology. In the following paper, we present an adapted topological watershed algorithm suited for a rapid and effective implementation on Shared Memory Parallel Machine (SMPM). The introduced algorithm allows a parallel watershed computing while preserving the given topology. No prior minima extraction is needed, nor the use of any sorting step or hierarchical queue. The strategy that guides the parallel watershed computing, labeled SDM-Strategy (equivalent to Split-Distributes and Merge), is also presented. Experimental analyses such as execution time, performance enhancement , cache consumption, efficiency and scalability are also presented and discussed.

• [hal-03111509] A minimal model for rotation invariant convolutional neural networks with angle prediction

  [...]

• [hal-02543994] On the asymptotic behaviour of the eigenvalue distribution of block correlation matrices of high-dimensional time series

  [...]

• [hal-03110182] Multi-Pedestrian Tracking on Crowded Scenes Using RGB Images

  [...]

• [hal-03110175] Rotation Equivariant CNNs for Image Classification

  [...]

• [hal-02495333] Sequence graphs realizations and ambiguity in language models

  Several natural language models rely on an assumption modeling each word context as a bag of words. We study the combinatorial implications of such assumption for the corresponding word or sentences representations. In particular , we present theoretical results concerning the family of sequence graphs, for which realizations yield equivalent representations given this assumption. Several combinatorial problems are presented, depending on three levels of generalisation (window size, graph orientation, and weights), and whether some of these are NP-complete is left opened. Based on these results, we also establish different algorithms, including a dynamic programming formulation, to count and explicit the different realizations of a sequence graph. This allows us to show that the bag of words assumption can induce an important number of sentences to have the same representations, even for relatively short context window sizes.

• [hal-03108479] A New Modelling Framework for Coarse-Grained Programmable Architectures

  Coarse-grained reconfigurable architectures (CGRA) are designed to deliver high-performance computing while drastically reducing the latency of the computing system. Although they are often highly domain-specifically optimized, they keep several levels of flexibility so that they can be reused. However, their reuse is generally limited due to the complexity of identifying the best allocation of new tasks into the hardware resources. Another limiting point is the complexity to produce a reliable performance analysis for each new implementation. To solve this problem, we propose to consider CGRA as a programmable, configuration-driven computing fabric, called Coarse-Grained Programmable Architecture (CGPA). We propose a new latency-based model to describe all hardware elements. We demonstrate how to implicitly model, with the help of latency's prediction, the heterogeneity of their material implementations. Our model provides the possibility to assess also the configuration cost, often neglected in other works. The design of the modelling framework allows it to become a part of a complete application mapping and scheduling chain, up to the automated generation of the execution context, thus maximizing the reusability of the given CGPA.

• [hal-03108451] Modelling and Mapping Framework for Coarse-GrainedProgrammable Architectures

  [...]

• [hal-03106734] Rotation Invariant Networks for Image Classification for HPC and Embedded Systems

  Convolutional Neural Network (CNNs) models’ size reduction has recently gained interest due to several advantages: energy cost reduction, embedded devices, and multi-core interfaces. One possible way to achieve model reduction is the usage of Rotation-invariant Convolutional Neural Networks because of the possibility of avoiding data augmentation techniques. In this work, we present the next step to obtain a general solution to endowing CNN architectures with the capability of classifying rotated objects and predicting the rotation angle without data-augmentation techniques. The principle consists of the concatenation of a representation mapping transforming rotation to translation and a shared weights predictor. This solution has the advantage of admitting different combinations of various basic, existing blocks. We present results obtained using a Gabor-filter bank and a ResNet feature backbone compared to previous other solutions. We also present the possibility to select between parallelizing the network in several threads for energy-aware High Performance Computing (HPC) applications or reducing the memory footprint for embedded systems. We obtain a competitive error rate on classifying rotated MNIST and outperform existing state-of-the-art results on CIFAR-10 when trained on up-right examples and validated on random orientations.

• [hal-03106592] Efficient graphical-processor-unit parallelization algorithm for computing Eigen values

  Several leading-edge applications such as pathology detection, biometric identification and face recognition are mainly based on blob and line detection. To address this problem, the Eigen value computing has been commonly employed due to its accuracy and robustness. However, the Eigen value computing requires a raised computational processing, an intensive memory data access and a data overlapping which involve higher execution times. To overcome these limitations, we propose in this paper a new parallel strategy to implement the Eigen value computing using a GPU. Our contributions are: (1) to optimize instruction scheduling in order to reduce the computation time, (2) to efficiently partition processing into blocks in order to increase the occupancy of streaming multiprocessors, (3) to provide efficient input data splitting on shared memory to take benefit from its lower access time, (4) and to propose new data management of shared memory so as to avoid access memory conflict and reduce memory bank accesses. Experimental results show that our proposed GPU parallel strategy for Eigen value computing achieves speedups of 27 compared to a multithreaded implementation, of 16 compared to a predefined function in the OpenCV library, and of 8 compared to a predefined function in the Cublas library, which are performed into a quad core multi-CPU platform. Next, our parallel strategy is evaluated through an Eigen value based method for retinal thick vessel segmentation which is essential for detecting ocular pathologies. The Eigen value computing is executed in 0.017 seconds, when using STARE database images. Accordingly, we have achieved real-time thick retinal vessel segmentation where average execution time is about 0.039 seconds.

• [hal-03105751] Rethinking Interactive Image Segmentation: Feature Space Annotation

  Despite the progress of interactive image segmentation methods, high-quality pixel-level annotation is still time-consuming and laborious --- a bottleneck for several deep learning applications. We take a step back to propose interactive and simultaneous segment annotation from multiple images guided by feature space projection and optimized by metric learning as the labeling progresses. This strategy is in stark contrast to existing interactive segmentation methodologies, which perform annotation in the image domain. We show that our approach can surpass the accuracy of state-of-the-art methods in foreground segmentation datasets: iCoSeg, DAVIS, and Rooftop. Moreover, it achieves 91.5\% accuracy in a known semantic segmentation dataset, Cityscapes, being 74.75 times faster than the original annotation procedure. The appendix presents additional qualitative results. Code and video demonstration will be released upon publication.

• [hal-03096130] Robust multimodal biometric authentication on IoT device through ear shape and arm gesture

  Nowadays, authentication is required for both physical access to buildings and internal access to computers and systems. Biometrics are one of the emerging technologies used to protect these highly sensitive structures. However, biometric systems based on a single trait enclose several problems such as noise sensitivity and vulnerability to spoof attacks. In this regard, we present in this paper a fully unobtrusive and robust multimodal authentication system that automatically authenticates a user by the way he/she answers his/her phone, after extracting ear and arm gesture biometric modalities from this single action. To deal the challenges facing ear and arm gesture authentication systems in real-world applications, we propose a new method based on image fragmentation that makes the ear recognition more robust in relation to occlusion. The ear feature extraction process has been made locally using Local Phase Quantization (LPQ) in order to get robustness with respect to pose and illumination variation. We also propose a set of four statistical metrics to extract features from arm gesture signals. The two modalities are combined on score-level using a weighted sum. In order to evaluate our contribution, we conducted a set of experiments to demonstrate the contribution of each of the two biometrics and the advantage of their fusion on the overall performance of the system. The multimodal biometric system achieves an equal error rate (EER) of 5.15%.

• [hal-02093170] Positive solutions for large random linear systems

  Consider a large linear system where $A_n$ is a $n\times n$ matrix with independent real standard Gaussian entries, $\boldsymbol{1}_n$ is a $n\times 1$ vector of ones and with unknown the $n\times 1$ vector $\boldsymbol{x}_n$ satisfying $$ \boldsymbol{x}_n = \boldsymbol{1}_n +\frac 1{\alpha_n\sqrt{n}} A_n \boldsymbol{x}_n\, . $$ We investigate the (componentwise) positivity of the solution $\boldsymbol{x}_n$ depending on the scaling factor $\alpha_n$ as the dimension $n$ goes to $\infty$. We prove that there is a sharp phase transition at the threshold $\alpha^*_n =\sqrt{2\log n}$: below the threshold ($\alpha_n\ll \sqrt{2\log n}$), $\boldsymbol{x}_n$ has negative components with probability tending to 1 while above ($\alpha_n\gg \sqrt{2\log n}$), all the vector's components are eventually positive with probability tending to 1. At the critical scaling $\alpha^*_n$, we provide a heuristics to evaluate the probability that $\boldsymbol{x}_n$ is positive. Such linear systems arise as solutions at equilibrium of large Lotka-Volterra systems of differential equations, widely used to describe large biological communities with interactions such as foodwebs for instance. In the domaine of positivity of the solution $\boldsymbol{x}_n$, that is when $\alpha_n\gg \sqrt{2\log n}$, we establish that the Lotka-Volterra system of differential equations whose solution at equilibrium is precisely $\x_n$ is stable in the sense that its jacobian $$ {\mathcal J}(\boldsymbol{x}_n) = \mathrm{diag}(\boldsymbol{x}_n)\left(-I_n + \frac {A_n}{\alpha_n\sqrt{n}}\right) $$ has all its eigenvalues with negative real part with probability tending to one. Our results shed a new light and complement the understanding of feasibility and stability issues for large biological communities with interaction.

• [hal-03093082] Positive Solutions for Large Random Linear Systems

  [...]

• [hal-03095169] Non-Hermitian random matrices with a variance profile (II): properties and examples

  For each $n$, let $A_n=(\sigma_{ij})$ be an $n\times n$ deterministic matrix and let $X_n=(X_{ij})$ be an $n\times n$ random matrix with i.i.d.\ centered entries of unit variance. In the companion article \cite{cook2018non}, we considered the empirical spectral distribution $\mu_n^Y$ of the rescaled entry-wise product \[ Y_n = \frac 1{\sqrt{n}} A_n\odot X_n = \left(\frac1{\sqrt{n}} \sigma_{ij}X_{ij}\right) \] and provided a deterministic sequence of probability measures $\mu_n$ such that the difference $\mu^Y_n - \mu_n$ converges weakly in probability to the zero measure. A key feature in \cite{cook2018non} was to allow some of the entries $\sigma_{ij}$ to vanish, provided that the standard deviation profiles $A_n$ satisfy a certain quantitative irreducibility property. In the present article, we provide more information on the sequence $(\mu_n)$, described by a family of \emph{Master Equations}. We consider these equations in important special cases %such as separable variance profiles $\sigma^2_{ij}=d_i \widetilde d_j$ and such as sampled variance profiles $\sigma^2_{ij} = \sigma^2\left(\frac in, \frac jn \right)$ where $(x,y)\mapsto \sigma^2(x,y)$ is a given function on $[0,1]^2$. Associated examples are provided where $\mu_n^Y$ converges to a genuine limit. We study $\mu_n$'s behavior at zero. % and provide examples where $\mu_n$'s density is bounded, blows up, or vanishes while an atom appears. As a consequence, we identify the profiles that yield the circular law. Finally, building upon recent results from Alt et al. \cite{alt2018local,alt2019location}, we prove that, except possibly at the origin, $\mu_n$ admits a positive density on the centered disc of radius $\sqrt{\rho(V_

• [hal-03094736] Accurate Strategy for Mixed Criticality Scheduling

  [...]

• [hal-01701455] Unbounded Largest Eigenvalue of Large Sample Covariance Matrices: Asymptotics, Fluctuations and Applications

  Given a large sample covariance matrix $ S_N=\frac 1n\Gamma_N^{1/2}Z_N Z_N^*\Gamma_N^{1/2}\, , $ where $Z_N$ is a $N\times n$ matrix with i.i.d. centered entries, and $\Gamma_N$ is a $N\times N$ deterministic Hermitian positive semidefinite matrix, we study the location and fluctuations of $\lambda_{\max}(S_N)$, the largest eigenvalue of $S_N$ as $N,n\to\infty$ and $Nn^{-1} \to r\in(0,\infty)$ in the case where the empirical distribution $\mu^{\Gamma_N}$ of eigenvalues of $\Gamma_N$ is tight (in $N$) and $\lambda_{\max}(\Gamma_N)$ goes to $+\infty$. These conditions are in particular met when $\mu^{\Gamma_N}$ weakly converges to a probability measure with unbounded support on $\mathbb{R}^+$. We prove that asymptotically $\lambda_{\max}(S_N)\sim \lambda_{\max}(\Gamma_N)$. Moreover when the $\Gamma_N$'s are block-diagonal, and the following {\em spectral gap condition} is assumed: $$ \limsup_{N\to\infty} \frac{\lambda_2(\Gamma_N)}{\lambda_{\max}(\Gamma_N)}<1, $$ where $\lambda_2(\Gamma_N)$ is the second largest eigenvalue of $\Gamma_N$, we prove Gaussian fluctuations for $\lambda_{\max}(S_N)/\lambda_{\max}(\Gamma_N)$ at the scale $\sqrt{n}$. In the particular case where $Z_N$ has i.i.d. Gaussian entries and $\Gamma_N$ is the $N\times N$ autocovariance matrix of a long memory Gaussian stationary process $({\mathcal X}_t)_{t\in\mathbb{Z}}$, the columns of $\Gamma_N^{1/2} Z_N$ can be considered as $n$ i.i.d. samples of the random vector $({\mathcal X}_1,\dots,{\mathcal X}_N)^\tran$. We then prove that $\Gamma_N$ is similar to a diagonal matrix which satisfies all the required assumptions of our theorems, hence our results apply to this case.

• [hal-03092473] A New Collaborative Clustering Approach for the Internet of Vehicles (CCA-IoV)

  [...]

• [hal-03091938] Toward an approach for securing IMS signaling and media planes

  [...]

• [hal-03091925] An efficient biometric-based continuous authentication scheme with HMM prehensile movements modeling

  Biometric is an emerging technique for user authentication thanks to its efficiency compared to the traditional methods, such as passwords and access-cards. However, most existing biometric authentication systems require the cooperation of users and provide only a login time authentication. To address these drawbacks, we propose in this paper a new, efficient continuous authentication scheme based on the newly biometric trait that still under development: prehensile movements. In this work, we model the movements through Hidden Markov Model-Universal Background Model (HMM-UBM) with continuous observations based on Gaussian Mixture Model (GMM). Unlike the literature, the gravity signal is included. The results of the experiments conducted on a public database HMOG and on a proprietary database, collected under uncontrolled conditions, have shown that prehensile movements are very promising. This new biometric feature will allow users to be authenticated continuously, passively and in real time.

• [hal-03091920] Impact of Mobility Models on Energy Consumption in Unmanned Aerial Ad-Hoc Network

  [...]

• [hal-02994752] Homotopic Digital Rigid Motion: An Optimization Approach on Cellular Complexes

  Topology preservation is a property of rigid motions in ${\mathbb R^2}$, but not in $\mathbb Z^2$. In this article, given a binary object $\mathsf X \subset \mathbb Z^2$ and a rational rigid motion ${\mathcal R}$, we propose a method for building a binary object $\mathsf X_{\mathcal R} \subset \mathbb Z^2$ resulting from the application of ${\mathcal R}$ on a binary object $\mathsf X$. Our purpose is to preserve the homotopy-type between $\mathsf X$ and $\mathsf X_{\mathcal R}$. To this end, we formulate the construction of $\mathsf X_{\mathcal R}$ from $\mathsf X$ as an optimization problem in the space of cellular complexes with the notion of collapse on complexes. More precisely, we define a cellular space $\mathbb H$ by superimposition of two cubical spaces $\mathbb F$ and $\mathbb G$ corresponding to the canonical Cartesian grid of $\mathbb Z^2$ where $\mathsf X$ is defined, and the Cartesian grid induced by the rigid motion ${\mathcal R}$, respectively. The object $\mathsf X_{\mathcal R}$ is then computed by building a homotopic transformation within the space $\mathbb H$, starting from the cubical complex in $\mathbb G$ resulting from the rigid motion of $\mathsf X$ with respect to ${\mathcal R}$ and ending at a complex fitting $\mathsf X_{\mathcal R}$ in $\mathbb F$ that can be embedded back into $\mathbb Z^2$.

• [hal-03086034] CHEARP: Chord-based Hierarchical Energy-Aware Routing Protocol for Wireless Sensor Networks

  Wireless Sensor networks (WSNs) are mostly deployed in hostile en- vironments, where nodes may do not have any information about their location. Hence, the designed routing protocols and applications have to function indepen- dently from the nodes location. Moreover, extending lifetime in such networks is a critical and challenging issue, since they consist of miniaturized energy-constrained devices. The motivation of this paper is to design an energy efficient location- independent routing protocol for data delivery in WSNs. Therefore, a Chord-based Hierarchical Energy-Aware Routing Protocol (CHEARP) is developed with the fo- cus on preserving the energy consumption. In contrary to the existing DHT-based protocols that interconnect nodes independently of their physical proximity, this pa- per proposes an approximate logical structure to the physical one, where the aim is to minimize the average paths’ length. Simulation results show that the proposed solution reduces the transmission Load, minimizes the transmission delay, and ex- tends the network longevity.

• [hal-02955867] Normalisation of 16th and 17th century texts in French and geographical named entity recognition

  Both statistical and rule-based methods for named entity recognition are quite sensitive to the type of language used in the analysed texts. Former studies have shown for example that it was harder to detect named entities in SMS or microblog messages where words are abridged or changed to lowercase. In this article, we focus on old French texts to evaluate the impact of manual and automatic normalization before applying five geographical named entity recognition tools, as well as an improved version of one of them, in order to help building maps displaying the locations mentioned in ancient texts. Our results show that manual normalisation leads to better results for all methods and that automatic normalisation performs differently depending on the tool used to extract geographical named entities, but with a significant improvement on most methods.

• [hal-03066187] SuperDeConFuse: A Supervised Deep Convolutional Transform based Fusion Framework for Financial Trading Systems

  This work proposes a supervised multi-channel time-series learning framework for financial stock trading. Although many deep learning models have recently been proposed in this domain, most of them treat the stock trading time-series data as 2-D image data, whereas its true nature is 1-D time-series data. Since the stock trading systems are multi-channel data, many existing techniques treating them as 1-D time-series data are not suggestive of any technique to effectively fusion the information carried by the multiple channels. To contribute towards both of these shortcomings, we propose an end-to-end supervised learning framework inspired by the previously established (unsupervised) convolution transform learning framework. Our approach consists of processing the data channels through separate 1-D convolution layers, then fusing the outputs with a series of fully-connected layers, and finally applying a softmax classification layer. The peculiarity of our framework, that we call SuperDeConFuse (SDCF), is that we remove the nonlinear activation located between the multi-channel convolution layers and the fully-connected layers, as well as the one located between the latter and the output layer. We compensate for this removal by introducing a suitable regularization on the aforementioned layer outputs and filters during the training phase. Specifically, we apply a logarithm determinant regularization on the layer filters to break symmetry and force diversity in the learnt transforms, whereas we enforce the non-negativity constraint on the layer outputs to mitigate the issue of dead neurons. This results in the effective learning of a richer set of features and filters with respect to a standard convolutional neural network. Numerical experiments confirm that the proposed model yields considerably better results than state-of-the-art deep learning techniques for the real-world problem of stock trading.

• [hal-03066160] Deeply Transformed Subspace Clustering

  Subspace clustering assumes that the data is separable into separate subspaces; this assumption may not always hold. For such cases, we assume that, even if the raw data is not separable into subspaces, one can learn a deep representation such that the learnt representation is separable into subspaces. To achieve the intended goal, we embed subspace clustering techniques (locally linear manifold clustering, sparse subspace clustering and low rank representation) into deep transform learning. The entire formulation is jointly learnt; giving rise to a new class of methods called deeply transformed subspace clustering (DTSC). To test the performance of the proposed techniques, benchmarking is performed on image clustering problems. Comparison with state-of-the-art clustering techniques shows that our formulation improves upon them.