Program 2020

The goal of this tutorial is to explain the bitcoin algorithm from the distributed computing perspective, precisely define the underlying double-payment problem, and present a simpler alternative to solve the problem without relying on consensus.

Rachid Guerraoui is professor in computer science at EPFL where he leads the laboratory of Distributed Computing. He worked in the past with Ecole des Mines de Paris, CEA Saclay, HP Labs in Palo Alto and MIT. He has been elected ACM fellow and professor of the College de France. He was awarded a Senior ERC Grant and a Google Focused Award.

Computing systems that make human sense of big data, usually called personalization systems or recommenders, and popularized by Amazon and Netflix, essentially help Internet users extracting information of interest to them. Building an operational recommender goes far beyond, especially in a world where data is not only big but also changes very fast. This tutorial will discuss system challenges to scale to a large number of users and a growing volume of fastly changing data to eventually provide real-time personalization.

Anne-Marie Kermarrec is Professor in Computer Science at EPFL since Jan. 2020. From 2015 to 2019, she was the CEO and co-founder of the Mediego Startup, leveraging her research work conducted during her ERC project. Before that she was with Inria, Microsoft Reasearch (UK) and Vrije Universiteit. She is an ACM fellow since 2016.

The Android operating system boasts 86% of the market share of the smartphone market. This astonishing widespread adoption is not without consequences. It comes in pairs with a dangerously fast spreading of malware across the Android ecosystem at an alarming rate. It turns out that detecting malicious software is a major key issue and represent a huge challenge. Accordingly, over the past decade, considerable ef- forts have been achieved to design various analysis techniques to detect malware. Nowadays huge amount of top notch anti- virus/malware systems claim to reach a high detection rate. Such a successful malware detection comes from the widely use of machine learning algorithms. However, malware complexity is ever continuously increasing either to exploit new threats and new vulnerabilities, or to simply try to circumvent these new generation of smart malware detection systems. In this work, we introduce a middleware Killer*Droid that enables to generate malware that bypass systematically the most advanced malware detection systems. In particular, we demonstrate that even the use of machine learning can not detect the next generation of malware. Our tool, Killer*Droid, is able to produce a new running malware from the combination of any existing malware and benign application without modifying its original behavior to bypass all existing antivirus. Precisely, we generated 75000 malwares to audit existing antiviruses and state of the art analysis solutions. We first show that almost none of the 60 audited antivirus is able to deal with these new malware samples. Finally, we demonstrate that this newly generated dataset drives down the detection rate of existing off-the-shelf malware detection approaches. Our approach makes obsolete traditional malware datasets usually used in the community for training smart malware detection tools. Further, we provide up to date datasets, and key insights on designing the next generation of anti malware systems.

David Bromberg has been a Professor in Distributed Computer Systems at the Université de Rennes (IRISA) since 2015. Prior to that, he was an Associate Professor at the University of Bordeaux and member of the LaBRI Software Engineering research group from 2008 to 2015. David holds a PhD from INRIA (2006) and an Habilitation to Head Research (HDR) from the University of Bordeaux (2014). His main interest lies in the scalability and programmability of complex distributed systems and software engineering applied to systems, distributed systems and network programming. David has authored over 30 peer-reviewed publications, and served on a number of program committees in his field.

Everything we do on the Web is collected, analyzed and used to build profiles on us. Our activities are collected and shared over many different services allowing a very detailed view of what we do. This is possible through various tracking techniques and the use of common tracking services that synchronize the user's activities. In this talk we will present the current status of tracking on the Web and describe how users are tracked through stateful techniques, such as cookies, and stateless techniques, such as browser fingerprinting.

Walter Rudametkin is an associate professor at the University of Lille and part of the Spirals team, a joint team between the CRIStAL laboratory and Inria. His work focuses on privacy on the Web, in particular, on studying the risks, uses and defenses of stateless tracking techniques, such as browser fingerprinting.

Big data has been very popular in recent years. It appears in many application domains, such as computational medicine, online advertisement, and recommendation systems, etc. Big data generates a lot values and capabilities for machine learning algorithms, but also creates many challenges. This lecture will cover the basic mechanisms of how machine learning approaches work.

Carole Frindel received the M.Sc. degree in computer vision from the Ecole Polytechnique de Montréal, Canada in 2005, M.Sc. degree in bioinformatics from the Institut National des Sciences Appliquées, France, in 2006 and the and the Ph.D. degree in electrical and computer engineering from the Institut National des Sciences Appliquées, France, in 2009. From 2009 to 2011, she worked in the industry for medical imaging software publishing groups. Since 2011, she is associate professor at the biomedical imaging research laboratory of the Institut National des Sciences Appliquées in Lyon. Her current interests are the applications of machine learning and deep learning to major health issues, including the prediction of pathologies and the contribution of iot in medicine.

The internet of the future will care about path lengths. A compelling case involves one-hop paths, which became possible thanks to direct communication technologies between nearby devices. While such a type of communication is quite common at small scales, their usage as a global offloading strategy still requires better knowledge of link capacity and more efficient consumption of radio resources. In this talk, we will identify some open issues and discuss a few recent contributions on the topic, including measurement and characterization of ephemerous links.

Marcelo Dias de Amorim is a CNRS Research Director and member of the LIP6 laboratory at Sorbonne Université. The focuses of his research are on understanding, designing, and evaluating interactive dynamic networks.

The requirements of large-scale distributed storage systems are ill-suited for classical error-correcting codes. For instance, even though Reed-Solomon codes, originally developed in 1960, have the greatest error-correction capability for a given storage overhead, they are the most expensive solution when a single transient failure requires decoding data. In this tutorial, I will discuss recent work on the capacity of distributed storage systems, and present the new coding techniques developed in the last 10 years in this setting.

Dr Hugues Mercier received the B.Sc. degree in mathematics from Université Laval, the M.Sc. degree in computer science from the Université de Montréal, and the Ph.D. degree in electrical and computer engineering from the University of British Columbia, Canada, in 2008. From 2008 to 2011, he was a postdoctoral research fellow at the Harvard School of Engineering and Applied Sciences, and at McGill University. Currently, he is a research associate at the Université de Neuchâtel in Switzerland. His current interests are the applications of coding theory, information theory, combinatorics, and algorithms to the study of communication networks. In his spare time, he does consulting on sports analytics for international sports federations.