Publications

2020

• Software Acceleration Techniques for High-speed Programmable Networks
  
  • Linguaglossa Leonardo
  , 2020, 6, pp.203-216. Network programmability has provided an effective approach to enable innovation in network systems. By replacing static, expensive middleboxes with equivalent pieces of software implementing the same functionality, operators can significantly reduce their CAPEX/OPEX expenditures, and engineers can rapidly design, test and deploy novel architectures and services, thus reducing the time-to-market for network applications. However, the flexibility provided by software solutions comes at a cost: purposespecific hardware has the clear advantage of optimized performance (i.e., throughput, latency) with respect to pure software-based solutions. The introduction of software acceleration techniques represented an essential step towards the increasing popularity of the SDN/NFV paradigm shift, by reducing the performance gap between hardware-based and software-based systems. Thanks to such techniques, modern software-networking solutions can operate at multi-10Gbps rate (up to hundreds of Gbps) on commodity servers equipped with regular COTS components. In this chapter, we cover the aspects related to software acceleration techniques in a bottom-up fashion: we first provide an overview of high-speed software networking on COTS architectures, and we then explore the evolution of softwarized networking by focusing on performance acceleration and the design space for high-speed programmable networks. (10.57620/CNIT-Report_06)
  DOI : 10.57620/CNIT-Report_06
• An Experimental Study of State-of-the-Art Entity Alignment Approaches
  
  • Zhao Xiang
  • Zeng Weixin
  • Tang Jiuyang
  • Wang​ Wei
  • Suchanek Fabian
  IEEE Transactions on Knowledge and Data Engineering, Institute of Electrical and Electronics Engineers, 2020. Entity alignment (EA) finds equivalent entities that are located in different knowledge graphs (KGs), which is an essential step to enhance the quality of KGs, and hence of significance to downstream applications (e.g., question answering and recommendation). Recent years have witnessed a rapid increase of EA approaches, yet the relative performance of them remains unclear, partly due to the incomplete empirical evaluations, as well as the fact that comparisons were carried out under different settings (i.e., datasets, information used as input, etc.). In this paper, we fill in the gap by conducting a comprehensive evaluation and detailed analysis of state-of-the-art EA approaches. We first propose a general EA framework that encompasses all the current methods, and then group existing methods into three major categories. Next, we judiciously evaluate these solutions on a wide range of use cases, based on their effectiveness, efficiency and robustness. Finally, we construct a new EA dataset to mirror the real-life challenges of alignment, which were largely overlooked by existing literature. This study strives to provide a clear picture of the strengths and weaknesses of current EA approaches, so as to inspire quality follow-up research. (10.1109/TKDE.2020.3018741)
  DOI : 10.1109/TKDE.2020.3018741
• Processing Simple Geometric Attributes with Autoencoders
  
  • Newson Alasdair
  • Almansa Andrés
  • Gousseau Yann
  • Ladjal Saïd
  Journal of Mathematical Imaging and Vision, Springer Verlag, 2020, 62 (3), pp.293-312. Image synthesis is a core problem in modern deep learning, and many recent architectures such as autoencoders and Generative Adversarial networks produce spectacular results on highly complex data, such as images of faces or landscapes. While these results open up a wide range of new, advanced synthesis applications, there is also a severe lack of theoretical understanding of how these networks work. This results in a wide range of practical problems, such as difficulties in training, the tendency to sample images with little or no variability, and generalisation problems. In this paper, we propose to analyse the ability of the simplest generative network, the autoencoder, to encode and decode two simple geometric attributes : size and position. We believe that, in order to understand more complicated tasks, it is necessary to first understand how these networks process simple attributes. For the first property, we analyse the case of images of centred disks with variable radii. We explain how the autoencoder projects these images to and from a latent space of smallest possible dimension, a scalar. In particular, we describe a closed-form solution to the decoding training problem in a network without biases, and show that during training, the network indeed finds this solution. We then investigate the best regularisation approaches which yield networks that generalise well. For the second property, position, we look at the encoding and decoding of Dirac delta functions, also known as `one-hot' vectors. We describe a hand-crafted filter that achieves encoding perfectly, and show that the network naturally finds this filter during training. We also show experimentally that the decoding can be achieved if the dataset is sampled in an appropriate manner. (10.1007/s10851-019-00924-w)
  DOI : 10.1007/s10851-019-00924-w
• Fractional Underdamped Langevin Dynamics: Retargeting SGD with Momentum under Heavy-Tailed Gradient Noise
  
  • Şimşekli Umut
  • Zhu Lingjiong
  • Teh Yee Whye
  • Gürbüzbalaban Mert
  , 2020. Stochastic gradient descent with momentum (SGDm) is one of the most popular optimization algorithms in deep learning. While there is a rich theory of SGDm for convex problems, the theory is considerably less developed in the context of deep learning where the problem is non-convex and the gradient noise might exhibit a heavy-tailed behavior, as empirically observed in recent studies. In this study, we consider a \emph{continuous-time} variant of SGDm, known as the underdamped Langevin dynamics (ULD), and investigate its asymptotic properties under heavy-tailed perturbations. Supported by recent studies from statistical physics, we argue both theoretically and empirically that the heavy-tails of such perturbations can result in a bias even when the step-size is small, in the sense that \emph{the optima of stationary distribution} of the dynamics might not match \emph{the optima of the cost function to be optimized}. As a remedy, we develop a novel framework, which we coin as \emph{fractional} ULD (FULD), and prove that FULD targets the so-called Gibbs distribution, whose optima exactly match the optima of the original cost. We observe that the Euler discretization of FULD has noteworthy algorithmic similarities with \emph{natural gradient} methods and \emph{gradient clipping}, bringing a new perspective on understanding their role in deep learning. We support our theory with experiments conducted on a synthetic model and neural networks.
• Optimizing Inner Product Masking Scheme by A Coding Theory Approach
  
  • Cheng Wei
  • Guilley Sylvain
  • Carlet Claude
  • Mesnager Sihem
  • Danger Jean-Luc
  IEEE Transactions on Information Forensics and Security, Institute of Electrical and Electronics Engineers, 2020, 16, pp.220-235. Masking is one of the most popular countermeasures to protect cryptographic implementations against side-channel analysis since it is provably secure and can be deployed at the algorithm level. To strengthen the original Boolean masking scheme, several works have suggested using schemes with high algebraic complexity. The Inner Product Masking (IPM) is one of those. In this paper, we propose a unified framework to quantitatively assess the side-channel security of the IPM in a coding-theoretic approach. Specifically, starting from the expression of IPM in a coded form, we use two defining parameters of the code to characterize its side-channel resistance. In order to validate the framework, we then connect it to two leakage metrics (namely signal-to-noise ratio and mutual information, from an information-theoretic aspect) and one typical attack metric (success rate, from a practical aspect) to build a firm foundation for our framework. As an application, our results provide ultimate explanations on the observations made by Balasch et al. at EUROCRYPT'15 and at ASIACRYPT'17, Wang et al. at CARDIS'16 and Poussier et al. at CARDIS'17 regarding the parameter effects in IPM, like higher security order in bounded moment model. Furthermore, we show how to systematically choose optimal codes (in the sense of a concrete security level) to optimize IPM by using this framework. Eventually, we present a simple but effective algorithm for choosing optimal codes for IPM, which is of special interest for designers when selecting optimal parameters for IPM. (10.1109/TIFS.2020.3009609)
  DOI : 10.1109/TIFS.2020.3009609
• Constructions of self-orthogonal codes from hulls of BCH codes and their parameters
  
  • Du Z.
  • Li C.
  • Mesnager Sihem
  IEEE Transactions on Information Theory, Institute of Electrical and Electronics Engineers, 2020.
• Recent results and problems on constructions of linear codes from cryptographic functions
  
  • Li N.
  • Mesnager Sihem
  Journal of Cryptography and Communications- Discrete Structures, Boolean Functions, and Sequences, 2020.
• Holarchic structures for decentralized deep learning: a performance analysis
  
  • Pournaras Evangelos
  • Yadhunathan Srivatsan
  • Diaconescu Ada
  Cluster Computing, Springer Verlag, 2020.
• Groove2Groove: One-Shot Music Style Transfer with Supervision from Synthetic Data
  
  • Cífka Ondřej
  • Şimşekli Umut
  • Richard Gael
  IEEE/ACM Transactions on Audio, Speech and Language Processing, Institute of Electrical and Electronics Engineers, 2020, 28, pp.2638-2650. Style transfer is the process of changing the style of an image, video, audio clip or musical piece so as to match the style of a given example. Even though the task has interesting practical applications within the music industry, it has so far received little attention from the audio and music processing community. In this paper, we present Groove2Groove, a one-shot style transfer method for symbolic music, focusing on the case of accompaniment styles in popular music and jazz. We propose an encoder-decoder neural network for the task, along with a synthetic data generation scheme to supply it with parallel training examples. This synthetic parallel data allows us to tackle the style transfer problem using end-to-end supervised learning, employing powerful techniques used in natural language processing. We experimentally demonstrate the performance of the model on style transfer using existing and newly proposed metrics, and also explore the possibility of style interpolation. (10.1109/TASLP.2020.3019642)
  DOI : 10.1109/TASLP.2020.3019642
• Artifical Intelligence and Pattern Recognition, Vision, Learning
  
  • Bloch Isabelle
  • Clouard Régis
  • Revenu Marinette
  • Sigaud Olivier
  , 2020, III, pp.337-364.
• Optical noise of dual-state lasing quantum dot lasers
  
  • Zhou Yueguang
  • Duan Jianan
  • Grillot Frederic
  • Wang Cheng
  IEEE Journal of Quantum Electronics, Institute of Electrical and Electronics Engineers, 2020, pp.1-1. (10.1109/JQE.2020.3026090)
  DOI : 10.1109/JQE.2020.3026090
• Lattice Codes for C-RAN Based Sectored Cellular Networks
  
  • Gelincik Samet
  • Rekaya-Ben Othman Ghaya
  , 2020.
• Precoding in Massive MU-MIMO Systems Based on New CSI Accuracy Indicator Reporting
  
  • Askri Aymen
  • Rekaya-Ben Othman Ghaya
  , 2020.
• Epitaxial quantum dot lasers on silicon with high thermal stability and strong resistance to optical feedback
  
  • Huang H.
  • Duan J.
  • Dong B.
  • Norman J.
  • Jung D.
  • Bowers J. E
  • Grillot F.
  APL Photonics, AIP Publishing LLC, 2020, 5 (1), pp.016103. (10.1063/1.5120029)
  DOI : 10.1063/1.5120029