#CNRSinsolite | Les visites insolites du CNRS

Au programme

Inscriptions sur le site du CNRS

https://www.cnrs.fr/fr/evenement/cnrsinsolite-les-visites-insolites-du-cnrs (région Ile-de-France hors Paris).

 

Dataflow Parallel Execution Models

Date: Friday, 17th of July, 2020 at 2:00 PM

Speaker: Stéphane Zuckerman

Title: Dataflow Parallel Execution Models

Abstract: Parallelism is now ubiquituous. Computer systems feature explicit parallel computing units, ranging from supercomputers with millions of compute units, down to embedded systems. However, most systems rely on the same basic abstract machine model: the Von Neumann abstract architecture, which is control-centric. Another possibility is to rely on data-centric models, such as data flow models of computation. This talk will introduce Dataflow models, from its origins to more modern aspects, such as the Codelet Model and its derivatives.

Séminaire ICI : Arnaud Duménil

Abstract

Polarization Dependent Loss (PDL) is expected to severely impact next-generation optical systems. It is a non-unitary effect that reduces the benefit of multiplexing in polarization in terms of channel capacity. In this presentation, I will present the associated loss of capacity and then provide solutions, both at the transmitter and at the receiver, that offer a resilience to this effect.

Short Bio

Arnaud Dumenil was born in Besançon, France in 1991. He received his Engineering Diploma from Centrale Marseille and a MSc in Telecommunications from DTU, Denmark, both in 2016. He then received a PhD degree in May 2020 from the Institut Polytechnique de Paris, in strong collaboration with Nokia Bell Labs in France. His thesis addressed optical communications with non-unitary effects. He just started a post-doctoral research at the ETIS laboratory focusing on short-blocklength error correcting codes for IoT applications.

Security in B5G low latency scenarios

Link to the article: https://arxiv.org/pdf/2001.07162.pdf

Abstract

With the emergence of URLLC and mMTC, corresponding low complexity and low latency security mechanisms are needed. Promising lightweight mechanisms include physical unclonable functions (PUF), secret key generation (SKG) at the physical layer and localization based authentication, as considered in this talk. We will demonstrate how physical layer security (PLS) allows building a new breed of low latency security schemes, such as zero-round-trip-time (0-RTT) resumption authentication protocols combining PUF and SKG processes. Furthermore, hybrid PLS and crypto schemes, such as authenticated encryption (AE) using SKG, will be introduced. We will conclude this talk with a discussion on future directions in 6G security.

Bio

Dr. Arsenia Chorti is an Associate Professor (MCF) at ETIS UMR8051 since Sept. 2017 and Head of the ICI team of ETIS. She obtained her PhD from Imperial College; from 2010 to 2012 she was a Research Fellow at Princeton University where she is currently a visiting researcher. She served as Senior Lecturer and Lecturer at the Universities of Middlessex and Essex in 2008-2009 and 2013-2017, respectively. Her research interests include PLS and wireless communications and has published more than 70 journals, book chapters and conference papers in these topics. She a member of the IEEE Future Networks Security Workgroup, the IEEE Teaching Awards Committee, the IEEE P1940 Standardization Workgroup and an Associate Editor of the IEEE Open Journal in Signal Processing.

Séminaire ETIS : Astrid Jourdan

Résumé

L’exposé porte sur l’analyse et la planification d’expériences numériques pour l’étude de gros codes de calcul, et plus particulièrement sur l’étude du lien entre les variables d’entrée et de sortie d’un simulateur.

La partie planification concerne la construction de plans d’expériences permettant de déterminer comment fixer les valeurs des variables d’entrée afin de récolter un maximum d’information en un minimum de simulations

La partie analyse est constituée de deux axes :

  • L’analyse de sensibilité qui permet de déterminer quelles variables d’entrée ont un réel impact sur la variable de sortie. L’objectif est de mieux comprendre la relation entre les entrées et la sortie, et de réduire la dimension d’un problème en éliminant les variables non actives.
  • L’ajustement de métamodèles (krigeage, SVR, réseaux de neurones…) afin de remplacer un gros code de calcul dans différentes analyses (optimisation, analyse de sensibilité…)

Initialement, ces travaux ont été appliqués aux simulateurs de l’industrie pétrolière. Depuis 5 ans, ces outils statistiques sont développés en transversalité avec le département informatique sur des méthodes d’optimisation type métaheuristiques.

L’exposé se terminera par une présentation d’axes de recherche interdisciplinaires développés dans le cadre de projets ou contrats industriels liés aux sciences environnementales.

Séminaire ETIS : Florent Devin

Titre du séminaire et orateur

Composition de service web.

Florent Devin, CY Tech.

Date et lieu

Jeudi 4 juin 2020, 13h30.

Visio-conférence.

Résumé

À travers une sélection de ses travaux de recherche, F. Devin, enseignant/chercheur en informatique, vous présentera la composition de services Web. Nous discuterons de services web, d’architecture logicielle et de composition. Cette présentation sera l’occasion de comprendre ses travaux et de nouer des contacts afin de permettre une collaboration recherche approfondie entre CY et CYTech.

Modélisation de la communication dans le réseau sans fil corporel (WBAN)

Abstract

Le réseau sans fil corporel (WBAN) a reçu beaucoup d’attention dans le domaine de la surveillance médicale en raison de sa commodité. Compte tenu de la sécurité de l’utilisateur, des économies d’énergie des appareils WBAN, de l’occlusion et l’atténuation des signaux du corps humain et des effets de la mobilité humaine sur la connexion entre les capteurs, une communication efficace et sécurisée reste un défi pour la communication dans intra WBAN.

Je vous présenterai le modèle de base du système WBAN que nous avons étudié et l’analyse des performances. À partir de cela, nous proposons un modèle basé sur une chaîne de Markov en temps continu représentant la communication de diffusion dans le WBAN. Ce modèle nous permet de calculer différents types de performances du système WBAN. De plus, nous étendons le système WBAN à un modèle théorique et nous avons étudié l’efficacité de la diffusion à l’aide du schéma d’étiquetage.

Security Specification and Evaluation in Pluridisciplinary Systems

Abstract

Pluridisciplinary systems are complex by controlling and integrating entities of various natures: physics, information, computational, network elements that are combined to execute a precise task. At this level of complexity, the main difficulty in analyzing this type of system relies in modeling each component separately as well as the way in which the various components inter-operate especially with the existing of an important number of connected low-energy nodes or human/social factor.

This talk aims to answer these problems by proposing a formal and practical framework where it is possible to model different aspects of components, their connectivity, and to assess how well they are secure. First, the talk details the modeling aspect and the formalization of these systems, and present the dedicated analysis approach. Then, it introduces a mechanism to specify and to constraint security in such system. Farther, I take this opportunity to present my ongoing research activities.

Short biography

Samir Ouchani received the PhD degree in 2013 from the department of Computer Science & Computer Engineering, Concordia University, Canada. Since 2017, he is enseignant-chercheur at CESI Engineering High School (Aix-en-Provence, France) and focuses on developing and applying formal methods as well as data mining to design, analyze, and harden security in large-scale systems. His research interests are: computer security, software engineering, formal methods, and data mining. Samir’s ongoing research activities look to develop techniques to strengthen security and detecting flaws on interdisciplinary systems such as: cyber-physical systems and inter-connected objects.

On the information bottleneck principle to understand generalization in deep learning

Info

  • Location: Zoom
  • Language: English or French
  • Keywords: machine learning; Information theory; deep learning; neural network; information bottleneck; generalization bounds
  • Everyone is welcome, whether or not you work on machine learning or information theory. The presentation will also include the basics of machine learning and information theory in order to understand the global idea.

Abstract

In this talk, I discuss recent approaches that explore the Information Bottleneck principle [1] (i.e., the trade-off between information compression and prediction quality, in the sense of the information theory) in order to understand generalization in deep learning architectures. Tishby and Noga [2], then extended by Shwartz-Ziv and Tishby [3], suggest that the goal of deep networks is to optimize the IB principle for each layer, according to the following discoveries: 1) most of the training is spent on compressing the input to efficient representation and not fitting the training labels. 2) The representation compression begins when the training errors become small. 3) The converged layers lie on or very close to the IB theoretical bounds. Finally, I discuss the criticism from Saxe et al. [4] which has shown that such approach does not work for recent deep network architectures (convolution, ReLU, residual connections, etc.), the answer from Naftali Tishby to the work of Saxe et al., the confirmation of Naftali Tishby’s point from the recent work from Noshard, Zeng and Hero [5] and finally the opened questions in generalization in deep networks.

[1] Naftali Tishby, Feranando C. Pereira, William Bialek. “The Information Bottleneck Method“. In proceedings of 37th Annual Allerton Conference on Communication, Control and Computing,1999.
[2] Naftali Tishby and Zaslavsky Noga. “Deep learning and the information bottleneck principle“. In IEEE Information Theory Workshop (ITW), 2015.
[3] Ravid Shwartz-Ziv and Naftali Tishby. “Opening the black box of deep neural networks via information“. arXiv preprint arXiv:1703.00810, 2017.
[4] Andrew M. Saxe, Yamini Bansal, Joel Dapello, Madhu Advani, Artemy Kolchinsky, Brendan D. Tracay, and David D. Cox. “On the Information Bottleneck Theory of Deep Learning”. In Proceedings of the International Conference on Representation Learning (ICLR), 2018
[5] Morteza Noshad, Yu Zeng, Alfred O. Hero. “Scalable mutual information estimation using dependence graphs.” In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2019.