Publications

2021

• Triple Sensing Current Margin for Maintainable MRAM Yield at Sub-100% Tunnel Magnetoresistance Ratio
  
  • Cai Hao
  • Han Menglin
  • Zhou Yongliang
  • Liu Bo
  • Naviner Lirida
  IEEE Transactions on Magnetics, Institute of Electrical and Electronics Engineers, 2021, 57 (2), pp.1-5. Spin transfer torque magnetic random access memory (STT-MRAM) creates significant breakthroughs as a proper candidate of next-generation non-volatile memory (NVM). Although STT-MRAM achieves high endurance, low access latency, and power consumption, the yield issue remains one of the critical concerns in high-density and large-scale MRAM array design. In this article, a novel triple-current margin sensing amplifier (TM-SA) is proposed for maintainable MRAM yield based on the current-mode SA and transmission gate switches. The sensing current margin of the proposed TM-SA is three times enlarged compared to traditional current mean (CM)-SA and resistance mean (RM)-SA. With a seriously degraded tunnel magnetoresistance (TMR) ratio (sub-100%, as low as 10%), the maximum voltage margin is 4.6 times of conventional CM-SA and five times of RM-SA. Monte-Carlo simulation shows that sensing failure probability can be greatly alleviated with the proposed TM-SA. The performance of TM-SA with respect to voltage margin can be further improved than that of CM-SA and RM-SA. (10.1109/TMAG.2020.3011614)
  DOI : 10.1109/TMAG.2020.3011614
• Security Analysis of Out-of-Band Device Pairing Protocols: A Survey
  
  • Khalfaoui Sameh
  • Leneutre Jean
  • Villard Arthur
  • Ma Jingxuan
  • Urien Pascal
  Wireless Communications and Mobile Computing, Hindawi Publishing Corporation, 2021, 2021, pp.1-30. Numerous secure device pairing (SDP) protocols have been proposed to establish a secure communication between unidentified IoT devices that have no preshared security parameters due to the scalability requirements imposed by the ubiquitous nature of the IoT devices. In order to provide the most user-friendly IoT services, the usability assessment has become the main requirement. Thus, the complete security analysis has been replaced by a sketch of a proof to partially validate the robustness of the proposal. The few existing formal or computational security verifications on the SDP schemes have been conducted based on the assessment of a wide variety of uniquely defined security properties. Therefore, the security comparison between these protocols is not feasible and there is a lack of a unified security analysis framework to assess these pairing techniques. In this paper, we survey a selection of secure device pairing proposals that have been formally or computationally verified. We present a systematic description of the protocol assumptions, the adopted verification model, and an assessment of the verification results. In addition, we normalize the used taxonomy in order to enhance the understanding of these security validations. Furthermore, we refine the adversary capabilities on the out-of-band channel by redefining the replay capability and by introducing a new notion of delay that is dependent on the protocol structure that is more adequate for the ad hoc pairing context. Also, we propose a classification of a number of out-of-band channels based on their security properties and under our refined adversary model. Our work motivates the future SDP protocol designer to conduct a formal or a computational security assessment to allow the comparability between these pairing techniques. Furthermore, it provides a realistic abstraction of the adversary capabilities on the out-of-band channel which improves the modeling of their security characteristics in the protocol verification tools. (10.1155/2021/8887472)
  DOI : 10.1155/2021/8887472
• Energy modeling of Hoeffding tree ensembles
  
  • García-Martín Eva
  • Bifet Albert
  • Lavesson Niklas
  Intelligent Data Analysis, IOS Press, 2021, 25 (1), pp.81--104. Energy consumption reduction has been an increasing trend in machine learning over the past few years due to its socio-ecological importance. In new challenging areas such as edge computing, energy consumption and predictive accuracy are key variables during algorithm design and implementation. State-of-the-art ensemble stream mining algorithms are able to create highly accurate predictions at a substantial energy cost. This paper introduces the nmin adaptation method to ensembles of Hoeffding tree algorithms, to further reduce their energy consumption without sacrificing accuracy. We also present extensive theoretical energy models of such algorithms, detailing their energy patterns and how nmin adaptation affects their energy consumption. We have evaluated the energy efficiency and accuracy of the nmin adaptation method on five different ensembles of Hoeffding trees under 11 publicly available datasets. The results show that we are able to reduce the energy consumption significantly, by 21% on average, affecting accuracy by less than one percent on average. (10.3233/IDA-194890)
  DOI : 10.3233/IDA-194890
• Efficient scheduling of applications onto cloud FPGAs
  
  • Bertolino Matteo
  , 2021. This thesis has been realized in Télécom Paris and it has been financed by Nokia Bell Labs France. It founds its motivations in the increasing usage of hardware accelerators such as FPGAs and their recent integration in modern cloud data center [1][2]. In some cases, servers and FPGAs are rented to users and the cost is related to the utilization time. Thus, offering a better sharing of FPGA pools would interest all stakeholders, namely cloud providers and users. We focus on scheduling and, in particular, we focus on makespan minimization of applications. The latter are assumed to be composed of several dependent tasks, whose features (i.e., dependencies, execution time, resource requirements, and so on) are known prior to their execution. With respect to the state of the art, we have sought to design an approach which is, at the same time, (i) general, (ii) fast and (iii) of high-quality. Indeed, several related works represent the applications and the architecture through simple models (e.g., the FPGA is often represented only with the amount of reconfigurable logic). We retain that such simple models may lead to unfeasible scheduling. Moreover, the vast majority of them is either based on slow and precise algorithms or on fast heuristics whose quality is far from the optimum. We therefore propose a scheduling solution [3] characterized by a good quality in terms of makespan while keeping the decision time in the order of tens of milliseconds for common applications. The main contributions of the thesis are a modelling proposal for FPGAs, the design of a heuristic which targets the makespan minimization and the evaluation of this heuristic on a synthetic benchmark of pseudo-randomly generated applications. Additionally, we have integrated this method to a model-driven engineering (MDE) tool to better support the early design of embedded systems. Finally, we propose several extensions to extend the approach to different architectures.References:[1] A. M. Caulfield et al., "A cloud-scale acceleration architecture," 2016 49th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO), Taipei, 2016, pp. 1-13, doi: 10.1109/MICRO.2016.7783710[2] Amazon Web Services Elastic Compute Cloud, https://docs.aws.amazon.com/ec2/index.html?nc2=h_ql_doc_ec2[3] Matteo Bertolino, Andrea Enrici, Renaud Pacalet, Ludovic Apvrille. Efficient Scheduling of FPGAs for Cloud Data Center Infrastructures. Euromicro DSD 2020, Aug 2020, Portorož, Slovenia. Proceedings will be published on IEEExplore, paper available on HAL of Télécom Paris: https://hal.telecom-paris.fr/hal-02894662v1 (10.70675/13d5944bzddb3z4ac7z95d6zd846ae5be8f2)
  DOI : 10.70675/13d5944bzddb3z4ac7z95d6zd846ae5be8f2
• Heavy-tailed nature of stochastic gradient descent in deep learning : theoretical and empirical analysis
  
  • Nguyen Thanh Huy
  , 2021. In this thesis, we are concerned with the Stochastic Gradient Descent (SGD) algorithm. Specifically, we perform theoretical and empirical analysis of the behavior of the stochastic gradient noise (GN), which is defined as the difference between the true gradient and the stochastic gradient, in deep neural networks. Based on these results, we bring an alternative perspective to the existing approaches for investigating SGD. The GN in SGD is often considered to be Gaussian for mathematical convenience. This assumption enables SGD to be studied as a stochastic differential equation (SDE) driven by a Brownian motion. We argue that the Gaussianity assumption might fail to hold in deep learning settings and hence render the Brownian motion-based analyses inappropriate. Inspired by non-Gaussian natural phenomena, we consider the GN in a more general context that suggests that the GN is better approximated by a "heavy-tailed" alpha-stable random vector. Accordingly, we propose to analyze SGD as a discretization of an SDE driven by a Lévy motion. Firstly, to justify the alpha-stable assumption, we conduct experiments on common deep learning scenarios and show that in all settings, the GN is highly non-Gaussian and exhibits heavy-tails. Secondly, under the heavy-tailed GN assumption, we provide a non-asymptotic analysis for the discrete-time dynamics SGD to converge to the global minimum in terms of suboptimality. Finally, we investigate the metastability nature of the SDE driven by Lévy motion that can then be exploited for clarifying the behavior of SGD, especially in terms of `preferring wide minima'. More precisely, we provide formal theoretical analysis where we derive explicit conditions for the step-size such that the metastability behavior of SGD, viewed as a discrete-time SDE, is similar to its continuous-time limit. We show that the behaviors of the two systems are indeed similar for small step-sizes and we describe how the error depends on the algorithm and problem parameters. We illustrate our metastability results with simulations on a synthetic model and neural networks. Our results open up a different perspective and shed more light on the view that SGD prefers wide minima. (10.70675/674d4873zb009z425cz9c8ezb792c820253d)
  DOI : 10.70675/674d4873zb009z425cz9c8ezb792c820253d
• Comparing Representations for Audio Synthesis Using Generative Adversarial Networks
  
  • Richard Gaël
  • Nistal Javier
  • Plattner Stefan
  , 2021, pp.161-165. —In this paper, we compare different audio signal representations, including the raw audio waveform and a variety of time-frequency representations, for the task of audio synthesis with Generative Adversarial Networks (GANs). We conduct the experiments on a subset of the NSynth dataset. The architecture follows the benchmark Progressive Growing Wasserstein GAN. We perform experiments both in a fully non-conditional manner as well as conditioning the network on the pitch information. We quantitatively evaluate the generated material utilizing standard metrics for assessing generative models, and compare training and sampling times. We show that complex-valued as well as the magnitude and Instantaneous Frequency of the ShortTime Fourier Transform achieve the best results, and yield fast generation and inversion times. The code for feature extraction, training and evaluating the model is available online. (10.23919/Eusipco47968.2020.9287799)
  DOI : 10.23919/Eusipco47968.2020.9287799
• Multi-Domain Image-to-Image Translation with Adaptive Inference Graph
  
  • Nguyen The-Phuc
  • Lathuilière Stéphane
  • Ricci Elisa
  , 2021. In this work, we address the problem of multi-domain image-to-image translation with particular attention paid to computational cost. In particular, current state of the art models require a large and deep model in order to handle the visual diversity of multiple domains. In a context of limited computational resources, increasing the network size may not be possible. Therefore, we propose to increase the network capacity by using an adaptive graph structure. At inference time, the network estimates its own graph by selecting specific sub-networks. Sub-network selection is implemented using Gumbel-Softmax in order to allow end-to-end training. This approach leads to an adjustable increase in number of parameters while preserving an almost constant computational cost. Our evaluation on two publicly available datasets of facial and painting images shows that our adaptive strategy generates better images with fewer artifacts than literature methods
• DR2S : Deep Regression with Region Selection for Camera Quality Evaluation
  
  • Tworski Marcelin
  • Lathuilière Stéphane
  • Belkarfa Salim
  • Fiandrotti Attilio
  • Cagnazzo Marco
  , 2020. In this work, we tackle the problem of estimating a camera capability to preserve fine texture details at a given lighting condition. Importantly, our texture preservation measurement should coincide with human perception. Consequently, we formulate our problem as a regression one and we introduce a deep convolutional network to estimate texture quality score. At training time, we use ground-truth quality scores provided by expert human annotators in order to obtain a subjective quality measure. In addition, we propose a region selection method to identify the image regions that are better suited at measuring perceptual quality. Finally, our experimental evaluation shows that our learning-based approach outperforms existing methods and that our region selection algorithm consistently improves the quality estimation.
• CARRADA Dataset: Camera and Automotive Radar with Range-Angle-Doppler Annotations
  
  • Ouaknine Arthur
  • Newson Alasdair
  • Rebut Julien
  • Tupin Florence
  • Perez Patrick
  2020 25th International Conference on Pattern Recognition (ICPR), 2021. High quality perception is essential for autonomous driving (AD) systems. To reach the accuracy and robustness that are required by such systems, several types of sensors must be combined. Currently, mostly cameras and laser scanners (lidar) are deployed to build a representation of the world around the vehicle. While radar sensors have been used for a long time in the automotive industry, they are still under-used for AD despite their appealing characteristics (notably, their ability to measure the relative speed of obstacles and to operate even in adverse weather conditions). To a large extent, this situation is due to the relative lack of automotive datasets with real radar signals that are both raw and annotated. In this work, we introduce CARRADA, a dataset of synchronized camera and radar recordings with rangeangle-Doppler annotations. We also present a semi-automatic annotation approach, which was used to annotate the dataset, and a radar semantic segmentation baseline, which we evaluate on several metrics. Both our code and dataset are available online. (10.1109/icpr48806.2021.9413181)
  DOI : 10.1109/icpr48806.2021.9413181
• CANU-ReID: A Conditional Adversarial Network for Unsupervised person Re-IDentification
  
  • Delorme Guillaume
  • Xu Yihong
  • Lathuilière Stéphane
  • Horaud Radu
  • Alameda-Pineda Xavier
  , 2021, pp.4428-4435. Unsupervised person re-ID is the task of identifying people on a target data set for which the ID labels are unavailable during training. In this paper, we propose to unify two trends in unsupervised person re-ID: clustering & fine-tuning and adversarial learning. On one side, clustering groups training images into pseudo-ID labels, and uses them to fine-tune the feature extractor. On the other side, adversarial learning is used, inspired by domain adaptation, to match distributions from different domains. Since target data is distributed across different camera viewpoints, we propose to model each camera as an independent domain, and aim to learn domain-independent features. Straightforward adversarial learning yields negative transfer, we thus introduce a conditioning vector to mitigate this undesirable effect. In our framework, the centroid of the cluster to which the visual sample belongs is used as conditioning vector of our conditional adversarial network, where the vector is permutation invariant (clusters ordering does not matter) and its size is independent of the number of clusters. To our knowledge, we are the first to propose the use of conditional adversar-ial networks for unsupervised person re-ID. We evaluate the proposed architecture on top of two state-of-the-art clustering-based unsupervised person re-identification (re-ID) methods on four different experimental settings with three different data sets and set the new state-of-the-art performance on all four of them. Our code and model will be made publicly available at https://team.inria.fr/perception/canu-reid/. (10.1109/ICPR48806.2021.9412431)
  DOI : 10.1109/ICPR48806.2021.9412431
• Learning Visual Voice Activity Detection with an Automatically Annotated Dataset
  
  • Guy Sylvain
  • Lathuilière Stéphane
  • Mesejo Pablo
  • Horaud Radu
  , 2021, pp.4851-4856. Visual voice activity detection (V-VAD) uses visual features to predict whether a person is speaking or not. VVAD is useful whenever audio VAD (A-VAD) is inefficient either because the acoustic signal is difficult to analyze or because it is simply missing. We propose two deep architectures for V-VAD, one based on facial landmarks and one based on optical flow. Moreover, available datasets, used for learning and for testing VVAD, lack content variability. We introduce a novel methodology to automatically create and annotate very large datasets inthe-wild – WildVVAD – based on combining A-VAD with face detection and tracking. A thorough empirical evaluation showsthe advantage of training the proposed deep V-VAD models with this dataset. (10.1109/ICPR48806.2021.9412884)
  DOI : 10.1109/ICPR48806.2021.9412884
• A New IoT Trust Model Based on TLS-SE and TLS-IM Secure Elements: A Blockchain Use Case
  
  • Urien Pascal
  , 2021, pp.1-2. (10.1109/CCNC49032.2021.9369485)
  DOI : 10.1109/CCNC49032.2021.9369485
• Machine Learning Detection for SMiShing Frauds
  
  • Msahli Mounira
  • Boukari Badr Eddine
  • Ravi Akshaya
  , 2021, pp.1-2. (10.1109/CCNC49032.2021.9369640)
  DOI : 10.1109/CCNC49032.2021.9369640
• Innovative TLS 1.3 Identity Module for Trusted IoT Device
  
  • Urien Pascal
  , 2021, pp.1-4. (10.1109/CCNC49032.2021.9369656)
  DOI : 10.1109/CCNC49032.2021.9369656
• A Survey on the Current Security Landscape of Intelligent Transportation Systems
  
  • Lamssaggad Ayyoub
  • Benamar Nabil
  • Hafid Abdelhakim Senhaji
  • Msahli Mounira
  IEEE Access, IEEE, 2021, pp.1-1. (10.1109/ACCESS.2021.3050038)
  DOI : 10.1109/ACCESS.2021.3050038
• Survey on Feature Transformation Techniques for Data Streams
  
  • Bahri Maroua
  • Bifet Albert
  • Maniu Silviu
  • Gomes Heitor Murilo
  , 2021, pp.4796-4802. Mining high-dimensional data streams poses a fundamental challenge to machine learning as the presence of high numbers of attributes can remarkably degrade any mining task’s performance. In the past several years, dimension reduction (DR) approaches have been successfully applied for different purposes (e.g., visualization). Due to their high-computational costs and numerous passes overlarge data, these approaches pose a hindrance when processing infinite data streams that are potentially high-dimensional. The latter increases the resource-usage of algorithms that could suffer from the curse of dimensionality. To cope with these issues, some techniques for incremental DR have been proposed. In this paper, we provide a survey on reduction approaches designed to handle data streams and highlight the key benefits of using these approaches for stream mining algorithms. (10.24963/ijcai.2020/668)
  DOI : 10.24963/ijcai.2020/668
• Approximate Inference and Learning of State Space Models with Laplace Noise
  
  • Neri Julian
  • Depalle Philippe
  • Badeau Roland
  IEEE Transactions on Signal Processing, Institute of Electrical and Electronics Engineers, 2021, 69, pp.3176 - 3189. State space models have been extensively applied to model and control dynamical systems in disciplines including neuroscience, target tracking, and audio processing. A common modeling assumption is that both the state and data noise are Gaussian because it simplifies the estimation of the system's state and model parameters. However, in many real-world scenarios where the noise is heavy-tailed or includes outliers, this assumption does not hold, and the performance of the model degrades. In this aper, we present a new approximate inference algorithm for state space models with Laplace-distributed multivariate data that is robust to a wide range of non-Gaussian noise. Exact inference is combined with an expectation propagation algorithm, leading to filtering and smoothing that outperforms existing approximate inference methods for Laplace-distributed data, while retaining a fast speed similar to the Kalman filter. Further, we present a maximum posterior expectation-maximization (EM) algorithm that learns the parameters of the model in an unsupervised way, automatically avoids over-fitting the data, and provides better model estimation than existing methods for the Gaussian model. The quality of the inference and learning algorithms are exemplified through a diverse set of experiments and an application to non-linear tracking of audio frequency. (10.1109/tsp.2021.3075146)
  DOI : 10.1109/tsp.2021.3075146
• Méta-apprentissage : classification de messages en catégories émotionnelles inconnues en entraînement
  
  • Guibon Gaël
  • Labeau Matthieu
  • Flamein Hélène
  • Lefeuvre Luce
  • Clavel Chloé
  , 2021, pp.199-208. Dans cet article nous reproduisons un scénario d’apprentissage selon lequel les données cibles ne sont pas accessibles et seules des données connexes le sont. Nous utilisons une approche par méta-apprentissage afin de déterminer si les méta-informations apprises à partir de messages issus de médias sociaux, finement annotés en émotions, peuvent produire de bonnes performances une fois utilisées sur des messages issus de conversations, étiquetés en émotions avec une granularité différente. Nous mettons à profit l’apprentissage sur quelques exemples (few-shot learning) pour la mise en place de ce scénario. Cette approche se montre efficace pour capturer les méta-informations d’un jeu d’étiquettes émotionnelles pour prédire des étiquettes jusqu’alors inconnues au modèle. Bien que le fait de varier le type de données engendre une baisse de performance, notre approche par méta-apprentissage atteint des résultats décents comparés au référentiel d’apprentissage supervisé.
• Sum-capacity of Uplink Multiband Satellite Communications with Nonlinear Impairments
  
  • Louchart Arthur
  • Ciblat Philippe
  • Poulliat Charly
  , 2021. A compact and closed-form expression of capacity is derived for a uplink multiband satellite system in the presence of nonlinear interference. The nonlinear effect comes from the satellite high-power amplifier modeled by a Volterra series expansion. The derivations reveal that the nonlinear interference can provide a constructive power contribution that could be used to increase the transmission rate. Consequently, decoders designed by viewing this interference as only an additional noise are suboptimal. Numerical results confirm this claim and also shows that an appropriate power allocation amongst the subbands may be of interest.
• Screening Rules and its Complexity for Active Set Identification
  
  • Ndiaye Eugene
  • Fercoq Olivier
  • Salmon Joseph
  Journal of Convex Analysis, Heldermann, 2021, 28 (4), pp.1053--1072. Screening rules were recently introduced as a technique for explicitly identifying active structures such as sparsity, in optimization problem arising in machine learning. This has led to new methods of acceleration based on a substantial dimension reduction. We show that screening rules stem from a combination of natural properties of subdifferential sets and optimality conditions, and can hence be understood in a unified way. Under mild assumptions, we analyze the number of iterations needed to identify the optimal active set for any converging algorithm. We show that it only depends on its convergence rate. (10.48550/arXiv.2009.02709)
  DOI : 10.48550/arXiv.2009.02709
• The Role of Digital Technologies in Responding to the Grand Challenges of the Natural Environment: The Windermere Accord
  
  • Blair Gordon
  • Bassett Richard
  • Bastin Louis
  • Beevers L.
  • Borrajo Garcia Maribel
  • Brown Mike
  • Dance Sarah L
  • Diaconescu Ada
  • Edwards Elizabeth
  • Ferrario Maria Angela
  • Fraser Robert
  • Harriet Fraser
  Patterns, Cell Press Elsevier, 2021.
• Infinite-dimensional gradient-based descent for alpha-divergence minimisation
  
  • Daudel Kamélia
  • Douc Randal
  • Portier François
  Annals of Statistics, Institute of Mathematical Statistics, 2021, 49 (4), pp.2250 - 2270. This paper introduces the $(\alpha, \Gamma)$-descent, an iterative algorithm which operates on measures and performs $\alpha$-divergence minimisation in a Bayesian framework. This gradient-based procedure extends the commonly-used variational approximation by adding a prior on the variational parameters in the form of a measure. We prove that for a rich family of functions $\Gamma$, this algorithm leads at each step to a systematic decrease in the $\alpha$-divergence and derive convergence results. Our framework recovers the Entropic Mirror Descent algorithm and provides an alternative algorithm that we call the Power Descent. Moreover, in its stochastic formulation, the $(\alpha, \Gamma)$-descent allows to optimise the mixture weights of any given mixture model without any information on the underlying distribution of the variational parameters. This renders our method compatible with many choices of parameters updates and applicable to a wide range of Machine Learning tasks. We demonstrate empirically on both toy and real-world examples the benefit of using the Power descent and going beyond the Entropic Mirror Descent framework, which fails as the dimension grows.
• A Latent Transformer for Disentangled Face Editing in Images and Videos
  
  • Yao Xu
  • Newson Alasdair
  • Gousseau Yann
  • Hellier Pierre
  , 2021, pp.13789-13798.
• Resolution of a Routing and Wavelength Assignment Problem by Independent Sets in Conflict Graphs
  
  • Hudry Olivier
  , 2021.
• Dual Optimization for Kolmogorov Model Learning Using Enhanced Gradient Descent
  
  • Duan Qiyou
  • Ghauch Hadi
  • Kim Taejoon
  IEEE Transactions on Signal Processing, Institute of Electrical and Electronics Engineers, 2021. Data representation techniques have made a substantial contribution to advancing data processing and machine learning (ML). Improving predictive power was the focus of previous representation techniques, which unfortunately perform rather poorly on the interpretability in terms of extracting underlying insights of the data. Recently, Kolmogorov model (KM) was studied, which is an interpretable and predictable representation approach to learning the underlying probabilistic structure of a set of random variables. The existing KM learning algorithms using semi-definite relaxation with randomization (SDRwR) or discrete monotonic optimization (DMO) have, however, limited utility to big data applications because they do not scale well computationally. In this paper, we propose a computationally scalable KM learning algorithm, based on the regularized dual optimization combined with enhanced gradient descent (GD) method. To make our method more scalable to large-dimensional problems, we propose two acceleration schemes, namely, eigenvalue decomposition (EVD) elimination strategy and proximal EVD algorithm. When applied to big data applications, it is demonstrated that the proposed method can achieve compatible training/prediction performance with significantly reduced computational complexity; roughly two orders of magnitude improvement in terms of the time overhead, compared to the existing KM learning algorithms. Furthermore, it is shown that the accuracy of logical relation mining for interpretability by using the proposed KM learning algorithm exceeds 80%.